{ "generated_at": "2026-05-14T01:04:26", "total_models": 113, "tested_count": 72, "tokens_per_call_assumption": { "input": 300, "output": 1500 }, "default_weights": { "quality": 0.5, "cost": 0.2, "tool_calling": 0.15, "speed": 0.075, "latency": 0.075 }, "subscriptions_catalog": { "ollama_cloud_pro": { "name": "Ollama Cloud", "plan": "Pro", "price_month_usd": 30, "url": "https://ollama.com/cloud", "notes": "Rate limit varía por modelo. Recomendado para uso a volumen mid (1-10k calls/día)." }, "xiaomi_standard": { "name": "Xiaomi MiMo Standard", "plan": "Standard", "price_month_usd": 14, "url": "https://mimo.xiaomi.com", "notes": "200M credits/mes. Off-peak 16-24 UTC = 0.8x consumption. Acceso a 8 modelos MiMo." }, "minimax_agent_pro": { "name": "MiniMax Agent Pro", "plan": "Agent Pro", "price_month_usd": 19, "url": "https://api.minimax.io", "notes": "Acceso a M2.7 highspeed + límites generosos para agentes (1k+ calls/día)." }, "anthropic_pro": { "name": "Anthropic Pro", "plan": "Pro", "price_month_usd": 20, "url": "https://www.anthropic.com/pricing", "notes": "Sub Anthropic Pro $20/mes. NO incluye API access (solo claude.ai web)." } }, "models": [ { "key": "groq-llama-4-scout", "id": "meta-llama/llama-4-scout-17b-16e-instruct", "name": "Llama 4 Scout 17B (Groq preview)", "tier": "cheap", "provider": "groq_direct", "open_source": true, "license": "Llama Community", "cost_input_per_M": 0.11, "cost_output_per_M": 0.34, "cost_per_1k_calls_usd": 0.543, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 157, "score_global": 7.69, "score_by_pillar": { "Contenido": 8.26, "Agentes": 7.79, "Coding": 8.15, "Razonamiento": 8.26 }, "score_by_suite": { "content_generation": 8.59, "tool_calling": 8.25, "task_management": 8.44, "code_generation": 8.63, "reasoning": 8.26, "summarization": 8.25, "presentation": 8.57, "startup_content": 8.4, "deep_reasoning": 8.1, "customer_support": 8.28, "structured_output": 8.37, "hallucination": 8.44, "creativity": 8.34, "string_precision": 8.05, "news_seo_writing": 7.49, "ocr_extraction": 7.73, "orchestration": 7.71, "multi_turn": 7.54, "policy_adherence": 8.04, "agent_capabilities": 6.7, "strategy": 8.38, "sales_outreach": 8.53, "translation": 8.29, "agent_long_horizon": 8.26, "niah_es": 6.91 }, "quality_avg": 7.7, "cost_score_avg": 8.32, "speed_score_avg": 7.81, "latency_score_avg": 5.87, "tool_calling_score_avg": 7.04, "judge_score_avg": 4.04, "tokens_per_second": 169.7, "latency_avg_s": 2.68, "total_input_tokens": 1518574, "total_output_tokens": 73662 }, { "key": "groq-llama-3.1-8b", "id": "llama-3.1-8b-instant", "name": "Llama 3.1 8B Instant (Groq)", "tier": "ultra_cheap", "provider": "groq_direct", "open_source": true, "license": "Llama Community", "cost_input_per_M": 0.05, "cost_output_per_M": 0.08, "cost_per_1k_calls_usd": 0.135, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 145, "score_global": 7.67, "score_by_pillar": { "Contenido": 8.27, "Agentes": 8.06, "Coding": 8.09, "Razonamiento": 7.8 }, "score_by_suite": { "content_generation": 8.65, "tool_calling": 8.64, "task_management": 8.46, "code_generation": 8.32, "reasoning": 8.11, "summarization": 8.14, "presentation": 8.34, "startup_content": 8.36, "deep_reasoning": 7.09, "customer_support": 7.77, "structured_output": 8.62, "hallucination": 8.32, "creativity": 8.49, "string_precision": 7.76, "news_seo_writing": 7.43, "ocr_extraction": 7.87, "orchestration": 7.83, "multi_turn": 8.0, "policy_adherence": 7.94, "agent_capabilities": 8.0, "strategy": 8.4, "sales_outreach": 8.43, "translation": 8.62, "agent_long_horizon": 8.21, "niah_es_lite": 6.7 }, "quality_avg": 7.33, "cost_score_avg": 8.72, "speed_score_avg": 8.02, "latency_score_avg": 6.3, "tool_calling_score_avg": 7.1, "judge_score_avg": 3.9, "tokens_per_second": 262.2, "latency_avg_s": 2.97, "total_input_tokens": 1533194, "total_output_tokens": 77242 }, { "key": "devstral", "id": "mistralai/devstral-small", "name": "Devstral Small", "tier": "ultra_cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.1, "cost_output_per_M": 0.3, "cost_per_1k_calls_usd": 0.48, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 223, "score_global": 7.52, "score_by_pillar": { "Contenido": 7.48, "Agentes": 7.21, "Coding": 8.09, "Razonamiento": 7.71 }, "score_by_suite": { "content_generation": 7.82, "tool_calling": 6.55, "task_management": 7.77, "code_generation": 8.14, "reasoning": 7.92, "summarization": 7.32, "presentation": 7.6, "startup_content": 7.8, "deep_reasoning": 7.64, "customer_support": 7.29, "structured_output": 7.77, "hallucination": 7.65, "creativity": 7.44, "string_precision": 8.69, "news_seo_writing": 6.77, "agent_capabilities": 7.07, "ocr_extraction": 7.11, "orchestration": 7.23, "multi_turn": 7.47, "policy_adherence": 7.54, "strategy": 7.65, "sales_outreach": 7.7, "translation": 8.29, "agent_long_horizon": 8.12, "niah_es": 7.24 }, "quality_avg": 7.89, "cost_score_avg": 7.57, "speed_score_avg": 8.47, "latency_score_avg": 5.42, "tool_calling_score_avg": 6.81, "judge_score_avg": 4.1, "tokens_per_second": 139.3, "latency_avg_s": 4.89, "total_input_tokens": 1644856, "total_output_tokens": 119582 }, { "key": "mistral-small-4", "id": "mistralai/mistral-small-2603", "name": "Mistral Small 4", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.15, "cost_output_per_M": 0.6, "cost_per_1k_calls_usd": 0.945, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 166, "score_global": 7.51, "score_by_pillar": { "Contenido": 7.91, "Agentes": 7.7, "Coding": 7.83, "Razonamiento": 7.87 }, "score_by_suite": { "content_generation": 8.22, "tool_calling": 7.42, "task_management": 8.28, "code_generation": 8.11, "reasoning": 8.17, "summarization": 7.6, "presentation": 7.63, "startup_content": 8.05, "deep_reasoning": 7.56, "customer_support": 8.3, "structured_output": 8.08, "hallucination": 8.03, "creativity": 8.08, "string_precision": 7.85, "news_seo_writing": 7.3, "ocr_extraction": 7.41, "orchestration": 6.87, "multi_turn": 7.35, "policy_adherence": 7.89, "agent_capabilities": 8.07, "strategy": 8.01, "sales_outreach": 8.02, "translation": 8.38, "agent_long_horizon": 7.68, "niah_es": 7.01 }, "quality_avg": 7.88, "cost_score_avg": 7.76, "speed_score_avg": 7.3, "latency_score_avg": 4.53, "tool_calling_score_avg": 7.07, "judge_score_avg": 4.13, "tokens_per_second": 81.8, "latency_avg_s": 9.93, "total_input_tokens": 1960762, "total_output_tokens": 140488 }, { "key": "groq-gpt-oss-20b", "id": "openai/gpt-oss-20b", "name": "GPT-OSS 20B (Groq)", "tier": "ultra_cheap", "provider": "groq_direct", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.075, "cost_output_per_M": 0.3, "cost_per_1k_calls_usd": 0.472, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 144, "score_global": 7.47, "score_by_pillar": { "Contenido": 8.25, "Agentes": 7.37, "Coding": 7.92, "Razonamiento": 7.78 }, "score_by_suite": { "content_generation": 8.82, "tool_calling": 7.32, "task_management": 7.78, "code_generation": 8.28, "reasoning": 8.28, "summarization": 8.0, "presentation": 8.11, "startup_content": 8.52, "deep_reasoning": 7.22, "customer_support": 7.0, "structured_output": 8.34, "hallucination": 8.21, "creativity": 8.27, "string_precision": 8.32, "news_seo_writing": 7.27, "ocr_extraction": 6.8, "orchestration": 6.41, "multi_turn": 8.0, "policy_adherence": 7.89, "agent_capabilities": 7.52, "strategy": 8.01, "sales_outreach": 8.57, "translation": 8.62, "agent_long_horizon": 7.98, "niah_es_lite": 6.66 }, "quality_avg": 7.1, "cost_score_avg": 8.47, "speed_score_avg": 9.26, "latency_score_avg": 5.89, "tool_calling_score_avg": 6.94, "judge_score_avg": 4.07, "tokens_per_second": 474.4, "latency_avg_s": 2.86, "total_input_tokens": 1367972, "total_output_tokens": 161997 }, { "key": "mimo-v2-omni-xiaomi", "id": "mimo-v2-omni", "name": "MiMo V2-Omni (Xiaomi direct)", "tier": "subscription", "provider": "xiaomi_direct", "open_source": true, "license": "MIT", "cost_input_per_M": 0.07, "cost_output_per_M": 0.07, "cost_per_1k_calls_usd": 0.126, "subscriptions": [ { "key": "xiaomi_standard", "name": "Xiaomi MiMo Standard", "plan": "Standard", "price_month_usd": 14, "url": "https://mimo.xiaomi.com", "notes": "200M credits/mes. Off-peak 16-24 UTC = 0.8x consumption. Acceso a 8 modelos MiMo." } ], "notes": "Mismo modelo que via OpenRouter — comparar provider stability", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 102, "score_global": 7.46, "score_by_pillar": { "Contenido": 7.8, "Agentes": 7.25, "Coding": 6.9, "Razonamiento": 7.35 }, "score_by_suite": { "content_generation": 8.4, "tool_calling": 7.27, "task_management": 8.54, "code_generation": 6.54, "reasoning": 8.36, "summarization": 6.51, "presentation": 8.5, "startup_content": 8.42, "deep_reasoning": 6.01, "customer_support": 6.93, "structured_output": 7.14, "hallucination": 8.1, "creativity": 8.16, "string_precision": 7.7, "news_seo_writing": 7.05, "ocr_extraction": 6.09, "orchestration": 6.09, "multi_turn": 7.58, "policy_adherence": 7.75, "agent_capabilities": 7.19, "strategy": 8.29, "sales_outreach": 8.38, "translation": 6.55, "agent_long_horizon": 8.17 }, "quality_avg": 7.27, "cost_score_avg": 8.84, "speed_score_avg": 8.93, "latency_score_avg": 2.09, "tool_calling_score_avg": 6.93, "judge_score_avg": 4.01, "tokens_per_second": 102.3, "latency_avg_s": 21.65, "total_input_tokens": 188123, "total_output_tokens": 170549 }, { "key": "mimo-v2.5", "id": "mimo-v2.5", "name": "MiMo V2.5 (Xiaomi)", "tier": "subscription", "provider": "xiaomi_direct", "open_source": false, "license": "Xiaomi Commercial", "cost_input_per_M": 0.07, "cost_output_per_M": 0.07, "cost_per_1k_calls_usd": 0.126, "subscriptions": [ { "key": "xiaomi_standard", "name": "Xiaomi MiMo Standard", "plan": "Standard", "price_month_usd": 14, "url": "https://mimo.xiaomi.com", "notes": "200M credits/mes. Off-peak 16-24 UTC = 0.8x consumption. Acceso a 8 modelos MiMo." } ], "notes": "All-in-one multimodal nativo, 1M context, lanzado 22 abril 2026", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 148, "score_global": 7.45, "score_by_pillar": { "Contenido": 7.97, "Agentes": 7.44, "Coding": 7.43, "Razonamiento": 7.26 }, "score_by_suite": { "content_generation": 8.29, "tool_calling": 7.31, "task_management": 6.75, "code_generation": 7.81, "reasoning": 6.78, "summarization": 7.64, "presentation": 8.2, "startup_content": 8.26, "deep_reasoning": 6.65, "customer_support": 8.08, "structured_output": 7.68, "hallucination": 8.08, "creativity": 8.07, "string_precision": 7.6, "news_seo_writing": 7.0, "ocr_extraction": 6.73, "orchestration": 7.38, "multi_turn": 7.49, "policy_adherence": 7.68, "agent_capabilities": 7.26, "strategy": 8.13, "sales_outreach": 8.49, "translation": 8.11, "agent_long_horizon": 8.01, "niah_es_lite": 7.05 }, "quality_avg": 7.63, "cost_score_avg": 8.47, "speed_score_avg": 7.8, "latency_score_avg": 1.98, "tool_calling_score_avg": 7.14, "judge_score_avg": 4.12, "tokens_per_second": 71.1, "latency_avg_s": 20.04, "total_input_tokens": 1691782, "total_output_tokens": 186765 }, { "key": "gemini-3.1-flash-lite", "id": "google/gemini-3.1-flash-lite-preview", "name": "Gemini 3.1 Flash Lite", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.25, "cost_output_per_M": 1.5, "cost_per_1k_calls_usd": 2.325, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 148, "score_global": 7.44, "score_by_pillar": { "Contenido": 7.92, "Agentes": 7.49, "Coding": 7.8, "Razonamiento": 7.89 }, "score_by_suite": { "content_generation": 8.25, "tool_calling": 7.84, "task_management": 8.25, "code_generation": 8.1, "reasoning": 8.13, "summarization": 7.55, "presentation": 7.77, "startup_content": 8.0, "deep_reasoning": 7.54, "customer_support": 7.16, "structured_output": 7.92, "hallucination": 8.18, "creativity": 8.08, "string_precision": 7.83, "news_seo_writing": 7.12, "ocr_extraction": 7.45, "orchestration": 6.83, "multi_turn": 7.82, "policy_adherence": 7.75, "agent_capabilities": 7.17, "strategy": 8.06, "sales_outreach": 8.43, "translation": 8.3, "agent_long_horizon": 7.57, "niah_es_lite": 6.78 }, "quality_avg": 7.82, "cost_score_avg": 7.36, "speed_score_avg": 7.59, "latency_score_avg": 4.98, "tool_calling_score_avg": 7.08, "judge_score_avg": 4.11, "tokens_per_second": 110.3, "latency_avg_s": 4.08, "total_input_tokens": 1348663, "total_output_tokens": 83980 }, { "key": "nemotron-nano", "id": "nvidia/nemotron-3-nano-30b-a3b", "name": "Nemotron 3 Nano 30B", "tier": "ultra_cheap", "provider": "openrouter", "open_source": true, "license": "NVIDIA Open", "cost_input_per_M": 0.05, "cost_output_per_M": 0.2, "cost_per_1k_calls_usd": 0.315, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 103, "score_global": 7.43, "score_by_pillar": { "Contenido": 7.85, "Agentes": 6.67, "Coding": 7.66, "Razonamiento": 7.74 }, "score_by_suite": { "content_generation": 8.0, "tool_calling": 6.84, "task_management": 7.95, "code_generation": 8.05, "reasoning": 7.89, "summarization": 7.5, "presentation": 8.15, "startup_content": 8.2, "deep_reasoning": 7.58, "customer_support": 6.4, "structured_output": 7.81, "hallucination": 7.72, "creativity": 7.97, "string_precision": 7.8, "news_seo_writing": 7.19, "ocr_extraction": 7.07, "orchestration": 5.72, "multi_turn": 7.17, "policy_adherence": 7.46, "agent_capabilities": 5.91, "strategy": 7.93, "sales_outreach": 7.95, "translation": 7.93, "agent_long_horizon": 7.5 }, "quality_avg": 7.79, "cost_score_avg": 8.65, "speed_score_avg": 8.21, "latency_score_avg": 1.65, "tool_calling_score_avg": 6.6, "judge_score_avg": 4.03, "tokens_per_second": 86.3, "latency_avg_s": 37.48, "total_input_tokens": 375017, "total_output_tokens": 310848 }, { "key": "mimo-v2.5-pro", "id": "mimo-v2.5-pro", "name": "MiMo V2.5-Pro (Xiaomi)", "tier": "subscription", "provider": "xiaomi_direct", "open_source": false, "license": "Xiaomi Commercial", "cost_input_per_M": 0.14, "cost_output_per_M": 0.14, "cost_per_1k_calls_usd": 0.252, "subscriptions": [ { "key": "xiaomi_standard", "name": "Xiaomi MiMo Standard", "plan": "Standard", "price_month_usd": 14, "url": "https://mimo.xiaomi.com", "notes": "200M credits/mes. Off-peak 16-24 UTC = 0.8x consumption. Acceso a 8 modelos MiMo." } ], "notes": "Flagship reasoning, agentic, 1M context, lanzado 22 abril 2026", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 103, "score_global": 7.42, "score_by_pillar": { "Contenido": 7.7, "Agentes": 7.3, "Coding": 7.53, "Razonamiento": 7.18 }, "score_by_suite": { "content_generation": 8.35, "tool_calling": 6.99, "task_management": 7.69, "code_generation": 8.04, "reasoning": 6.59, "summarization": 7.66, "presentation": 8.09, "startup_content": 8.23, "deep_reasoning": 6.81, "customer_support": 7.31, "structured_output": 7.54, "hallucination": 7.87, "creativity": 7.47, "string_precision": 7.26, "news_seo_writing": 7.01, "ocr_extraction": 7.42, "orchestration": 7.44, "multi_turn": 7.67, "policy_adherence": 7.28, "agent_capabilities": 6.89, "strategy": 7.82, "sales_outreach": 8.45, "translation": 6.41, "agent_long_horizon": 7.16 }, "quality_avg": 7.65, "cost_score_avg": 8.83, "speed_score_avg": 6.97, "latency_score_avg": 1.44, "tool_calling_score_avg": 7.11, "judge_score_avg": 4.13, "tokens_per_second": 49.1, "latency_avg_s": 31.75, "total_input_tokens": 209698, "total_output_tokens": 164371 }, { "key": "mimo-v2-flash", "id": "xiaomi/mimo-v2-flash", "name": "MiMo-V2-Flash", "tier": "ultra_cheap", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 0.09, "cost_output_per_M": 0.29, "cost_per_1k_calls_usd": 0.462, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 103, "score_global": 7.41, "score_by_pillar": { "Contenido": 7.86, "Agentes": 6.6, "Coding": 7.59, "Razonamiento": 7.79 }, "score_by_suite": { "content_generation": 8.21, "tool_calling": 6.78, "task_management": 8.1, "code_generation": 8.05, "reasoning": 8.19, "summarization": 7.7, "presentation": 7.96, "startup_content": 7.93, "deep_reasoning": 7.53, "customer_support": 5.28, "structured_output": 7.56, "hallucination": 7.73, "creativity": 7.9, "string_precision": 7.42, "news_seo_writing": 7.12, "ocr_extraction": 7.45, "orchestration": 5.88, "multi_turn": 7.67, "policy_adherence": 7.41, "agent_capabilities": 5.83, "strategy": 7.96, "sales_outreach": 8.22, "translation": 8.11, "agent_long_horizon": 7.55 }, "quality_avg": 7.72, "cost_score_avg": 8.81, "speed_score_avg": 6.25, "latency_score_avg": 2.46, "tool_calling_score_avg": 6.87, "judge_score_avg": 4.14, "tokens_per_second": 54.2, "latency_avg_s": 33.67, "total_input_tokens": 231731, "total_output_tokens": 129827 }, { "key": "mimo-v2-pro-xiaomi", "id": "mimo-v2-pro", "name": "MiMo V2-Pro (Xiaomi direct)", "tier": "subscription", "provider": "xiaomi_direct", "open_source": true, "license": "MIT", "cost_input_per_M": 0.07, "cost_output_per_M": 0.07, "cost_per_1k_calls_usd": 0.126, "subscriptions": [ { "key": "xiaomi_standard", "name": "Xiaomi MiMo Standard", "plan": "Standard", "price_month_usd": 14, "url": "https://mimo.xiaomi.com", "notes": "200M credits/mes. Off-peak 16-24 UTC = 0.8x consumption. Acceso a 8 modelos MiMo." } ], "notes": "Mismo modelo que via OpenRouter — comparar provider stability", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 103, "score_global": 7.39, "score_by_pillar": { "Contenido": 7.87, "Agentes": 7.28, "Coding": 7.16, "Razonamiento": 6.74 }, "score_by_suite": { "content_generation": 8.14, "tool_calling": 6.67, "task_management": 8.08, "code_generation": 8.28, "reasoning": 5.15, "summarization": 7.66, "presentation": 8.18, "startup_content": 8.23, "deep_reasoning": 7.24, "customer_support": 7.39, "structured_output": 7.46, "hallucination": 7.85, "creativity": 7.94, "string_precision": 7.14, "news_seo_writing": 7.07, "ocr_extraction": 6.05, "orchestration": 7.5, "multi_turn": 7.14, "policy_adherence": 7.22, "agent_capabilities": 7.16, "strategy": 6.23, "sales_outreach": 8.24, "translation": 7.73, "agent_long_horizon": 7.7 }, "quality_avg": 7.61, "cost_score_avg": 8.86, "speed_score_avg": 6.5, "latency_score_avg": 1.26, "tool_calling_score_avg": 6.96, "judge_score_avg": 4.14, "tokens_per_second": 44.8, "latency_avg_s": 36.83, "total_input_tokens": 248425, "total_output_tokens": 165247 }, { "key": "gpt-oss-120b-cloud", "id": "gpt-oss:120b-cloud", "name": "GPT-OSS 120B (Ollama Cloud)", "tier": "cloud_ollama", "provider": "ollama_cloud", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [ { "key": "ollama_cloud_pro", "name": "Ollama Cloud", "plan": "Pro", "price_month_usd": 30, "url": "https://ollama.com/cloud", "notes": "Rate limit varía por modelo. Recomendado para uso a volumen mid (1-10k calls/día)." } ], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 148, "score_global": 7.37, "score_by_pillar": { "Contenido": 8.07, "Agentes": 6.9, "Coding": 7.67, "Razonamiento": 7.8 }, "score_by_suite": { "content_generation": 8.32, "tool_calling": 7.12, "task_management": 6.67, "code_generation": 8.16, "reasoning": 8.29, "summarization": 7.76, "presentation": 8.11, "startup_content": 8.31, "deep_reasoning": 7.69, "customer_support": 6.31, "structured_output": 7.82, "hallucination": 7.41, "creativity": 8.09, "string_precision": 7.65, "news_seo_writing": 7.56, "ocr_extraction": 7.21, "orchestration": 6.33, "multi_turn": 7.41, "policy_adherence": 7.44, "agent_capabilities": 7.07, "strategy": 7.91, "sales_outreach": 8.28, "translation": 8.08, "agent_long_horizon": 8.6, "niah_es_lite": 6.65 }, "quality_avg": 7.15, "cost_score_avg": 10.0, "speed_score_avg": 7.47, "latency_score_avg": 2.79, "tool_calling_score_avg": 6.87, "judge_score_avg": 4.04, "tokens_per_second": 67.7, "latency_avg_s": 14.3, "total_input_tokens": 1465863, "total_output_tokens": 180251 }, { "key": "gemini-flash", "id": "google/gemini-2.5-flash", "name": "Gemini 2.5 Flash", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.3, "cost_output_per_M": 2.5, "cost_per_1k_calls_usd": 3.84, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 135, "score_global": 7.36, "score_by_pillar": { "Razonamiento": 7.37, "Agentes": 6.81, "Coding": 7.79, "Contenido": 7.53 }, "score_by_suite": { "deep_reasoning": 7.29, "customer_support": 5.85, "structured_output": 7.41, "hallucination": 7.55, "creativity": 7.26, "string_precision": 8.38, "news_seo_writing": 7.26, "content_generation": 7.9, "tool_calling": 6.63, "task_management": 7.57, "code_generation": 7.5, "reasoning": 7.43, "summarization": 7.56, "presentation": 7.48, "startup_content": 7.68, "ocr_extraction": 7.19, "orchestration": 7.28, "multi_turn": 7.1, "policy_adherence": 7.35, "agent_capabilities": 6.89, "strategy": 7.27, "sales_outreach": 8.03, "translation": 7.94, "agent_long_horizon": 7.34 }, "quality_avg": 7.87, "cost_score_avg": 7.28, "speed_score_avg": 8.61, "latency_score_avg": 3.6, "tool_calling_score_avg": 6.69, "judge_score_avg": 4.13, "tokens_per_second": 115.8, "latency_avg_s": 7.88, "total_input_tokens": 223199, "total_output_tokens": 154266 }, { "key": "groq-llama-3.3-70b", "id": "llama-3.3-70b-versatile", "name": "Llama 3.3 70B (Groq)", "tier": "cheap", "provider": "groq_direct", "open_source": true, "license": "Llama Community", "cost_input_per_M": 0.59, "cost_output_per_M": 0.79, "cost_per_1k_calls_usd": 1.362, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 169, "score_global": 7.36, "score_by_pillar": { "Contenido": 8.06, "Agentes": 7.6, "Coding": 7.88, "Razonamiento": 7.84 }, "score_by_suite": { "content_generation": 8.35, "tool_calling": 6.94, "task_management": 8.13, "code_generation": 8.01, "reasoning": 8.02, "summarization": 7.64, "presentation": 7.97, "startup_content": 8.07, "deep_reasoning": 7.54, "customer_support": 7.85, "structured_output": 8.29, "hallucination": 8.07, "creativity": 8.26, "string_precision": 7.62, "news_seo_writing": 7.29, "ocr_extraction": 7.76, "orchestration": 7.23, "multi_turn": 7.69, "policy_adherence": 7.74, "agent_capabilities": 7.74, "strategy": 8.05, "sales_outreach": 8.5, "translation": 8.58, "agent_long_horizon": 7.91, "niah_es": 6.27 }, "quality_avg": 7.72, "cost_score_avg": 7.13, "speed_score_avg": 7.77, "latency_score_avg": 5.67, "tool_calling_score_avg": 7.09, "judge_score_avg": 4.0, "tokens_per_second": 172.9, "latency_avg_s": 3.81, "total_input_tokens": 1953899, "total_output_tokens": 115755 }, { "key": "gemini-flash-lite", "id": "google/gemini-2.5-flash-lite", "name": "Gemini 2.5 Flash Lite", "tier": "ultra_cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.1, "cost_output_per_M": 0.4, "cost_per_1k_calls_usd": 0.63, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 155, "score_global": 7.34, "score_by_pillar": { "Contenido": 7.47, "Agentes": 6.76, "Coding": 7.89, "Razonamiento": 7.38 }, "score_by_suite": { "content_generation": 7.51, "tool_calling": 6.25, "task_management": 7.5, "code_generation": 7.62, "reasoning": 7.43, "summarization": 7.43, "presentation": 7.42, "startup_content": 7.54, "deep_reasoning": 7.27, "customer_support": 6.17, "structured_output": 7.71, "hallucination": 7.63, "creativity": 7.25, "string_precision": 8.48, "news_seo_writing": 7.17, "agent_capabilities": 6.51, "ocr_extraction": 7.2, "orchestration": 6.76, "multi_turn": 7.6, "policy_adherence": 7.67, "strategy": 7.24, "sales_outreach": 8.04, "translation": 8.23 }, "quality_avg": 7.79, "cost_score_avg": 7.25, "speed_score_avg": 9.06, "latency_score_avg": 4.48, "tool_calling_score_avg": 6.42, "judge_score_avg": 4.15, "tokens_per_second": 170.9, "latency_avg_s": 3.94, "total_input_tokens": 40134, "total_output_tokens": 129841 }, { "key": "gemma-4-31b", "id": "google/gemma-4-31b-it", "name": "Gemma 4 31B", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.3, "cost_output_per_M": 0.6, "cost_per_1k_calls_usd": 0.99, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 119, "score_global": 7.3, "score_by_pillar": { "Contenido": 7.4, "Agentes": 7.07, "Coding": 7.31, "Razonamiento": 7.53 }, "score_by_suite": { "content_generation": 7.05, "tool_calling": 7.06, "startup_content": 7.57, "task_management": 7.76, "code_generation": 7.7, "reasoning": 7.99, "summarization": 7.38, "presentation": 7.57, "deep_reasoning": 7.06, "customer_support": 7.5, "structured_output": 7.65, "hallucination": 7.48, "creativity": 7.89, "string_precision": 7.38, "news_seo_writing": 6.95, "ocr_extraction": 6.64, "orchestration": 6.05, "multi_turn": 7.36, "policy_adherence": 7.06, "agent_capabilities": 7.09, "strategy": 7.59, "sales_outreach": 7.81, "translation": 7.46, "agent_long_horizon": 7.34 }, "quality_avg": 8.19, "cost_score_avg": 8.39, "speed_score_avg": 3.61, "latency_score_avg": 1.82, "tool_calling_score_avg": 7.23, "judge_score_avg": 4.17, "tokens_per_second": 22.1, "latency_avg_s": 139.35, "total_input_tokens": 293775, "total_output_tokens": 129546 }, { "key": "nim-gemma-4-31b", "id": "google/gemma-4-31b-it", "name": "Gemma 4 31B (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Gemma Terms", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Mismo modelo que via OpenRouter (10 runs parciales) - completar via NIM gratis", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 119, "score_global": 7.3, "score_by_pillar": { "Contenido": 7.4, "Agentes": 7.07, "Coding": 7.31, "Razonamiento": 7.53 }, "score_by_suite": { "content_generation": 7.05, "tool_calling": 7.06, "startup_content": 7.57, "task_management": 7.76, "code_generation": 7.7, "reasoning": 7.99, "summarization": 7.38, "presentation": 7.57, "deep_reasoning": 7.06, "customer_support": 7.5, "structured_output": 7.65, "hallucination": 7.48, "creativity": 7.89, "string_precision": 7.38, "news_seo_writing": 6.95, "ocr_extraction": 6.64, "orchestration": 6.05, "multi_turn": 7.36, "policy_adherence": 7.06, "agent_capabilities": 7.09, "strategy": 7.59, "sales_outreach": 7.81, "translation": 7.46, "agent_long_horizon": 7.34 }, "quality_avg": 8.19, "cost_score_avg": 8.39, "speed_score_avg": 3.61, "latency_score_avg": 1.82, "tool_calling_score_avg": 7.23, "judge_score_avg": 4.17, "tokens_per_second": 22.1, "latency_avg_s": 139.35, "total_input_tokens": 293775, "total_output_tokens": 129546 }, { "key": "gpt-5.4-mini", "id": "gpt-5.4-mini", "name": "GPT-5.4 Mini", "tier": "cheap", "provider": "openai_direct", "open_source": false, "license": "", "cost_input_per_M": 0.5, "cost_output_per_M": 1.5, "cost_per_1k_calls_usd": 2.4, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 155, "score_global": 7.23, "score_by_pillar": { "Contenido": 7.16, "Agentes": 6.86, "Coding": 7.76, "Razonamiento": 7.33 }, "score_by_suite": { "content_generation": 7.64, "tool_calling": 6.71, "task_management": 7.03, "code_generation": 7.46, "reasoning": 7.01, "summarization": 7.15, "presentation": 7.54, "startup_content": 6.75, "deep_reasoning": 7.29, "customer_support": 6.51, "structured_output": 7.36, "hallucination": 7.69, "creativity": 6.57, "string_precision": 8.49, "news_seo_writing": 7.09, "agent_capabilities": 6.43, "ocr_extraction": 7.13, "orchestration": 7.4, "multi_turn": 7.28, "policy_adherence": 7.54, "strategy": 7.43, "sales_outreach": 7.98, "translation": 7.72 }, "quality_avg": 7.37, "cost_score_avg": 7.39, "speed_score_avg": 9.1, "latency_score_avg": 4.03, "tool_calling_score_avg": 7.07, "judge_score_avg": 4.15, "tokens_per_second": 117.9, "latency_avg_s": 5.02, "total_input_tokens": 37650, "total_output_tokens": 108242 }, { "key": "devstral-medium", "id": "mistralai/devstral-medium", "name": "Devstral Medium", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.4, "cost_output_per_M": 2.0, "cost_per_1k_calls_usd": 3.12, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 88, "score_global": 7.21, "score_by_pillar": { "Contenido": 7.51, "Agentes": 6.79, "Coding": 7.44, "Razonamiento": 7.26 }, "score_by_suite": { "content_generation": 7.72, "tool_calling": 6.3, "task_management": 7.84, "code_generation": 7.59, "reasoning": 7.37, "summarization": 6.92, "presentation": 7.86, "startup_content": 7.72, "deep_reasoning": 7.16, "customer_support": 6.53, "structured_output": 7.51, "hallucination": 7.29, "creativity": 7.79, "string_precision": 7.63, "news_seo_writing": 6.77, "ocr_extraction": 7.02, "orchestration": 6.88, "multi_turn": 6.43, "policy_adherence": 7.16, "agent_capabilities": 6.67, "strategy": 7.33, "sales_outreach": 7.87 }, "quality_avg": 7.89, "cost_score_avg": 7.98, "speed_score_avg": 6.11, "latency_score_avg": 2.51, "tool_calling_score_avg": 6.6, "judge_score_avg": 4.07, "tokens_per_second": 60.2, "latency_avg_s": 23.72, "total_input_tokens": 27995, "total_output_tokens": 45938 }, { "key": "grok-4.1-fast", "id": "x-ai/grok-4.1-fast", "name": "Grok 4.1 Fast", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.2, "cost_output_per_M": 0.5, "cost_per_1k_calls_usd": 0.81, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 148, "score_global": 7.21, "score_by_pillar": { "Contenido": 7.67, "Agentes": 7.51, "Coding": 7.7, "Razonamiento": 7.81 }, "score_by_suite": { "content_generation": 7.75, "tool_calling": 7.91, "task_management": 7.91, "code_generation": 7.84, "reasoning": 8.06, "summarization": 7.55, "presentation": 7.94, "startup_content": 7.97, "deep_reasoning": 7.7, "customer_support": 7.15, "structured_output": 7.92, "hallucination": 7.94, "creativity": 7.72, "string_precision": 7.92, "news_seo_writing": 7.01, "ocr_extraction": 7.14, "orchestration": 7.39, "multi_turn": 7.4, "policy_adherence": 7.41, "agent_capabilities": 7.53, "strategy": 7.63, "sales_outreach": 7.78, "translation": 7.91, "agent_long_horizon": 7.41, "niah_es_lite": 6.29 }, "quality_avg": 7.53, "cost_score_avg": 7.66, "speed_score_avg": 8.84, "latency_score_avg": 2.29, "tool_calling_score_avg": 7.14, "judge_score_avg": 4.19, "tokens_per_second": 104.6, "latency_avg_s": 21.37, "total_input_tokens": 1366100, "total_output_tokens": 220481 }, { "key": "nim-qwen3-next-instruct", "id": "qwen/qwen3-next-80b-a3b-instruct", "name": "Qwen 3-Next 80B Instruct (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Próxima generación Qwen — pendiente desde Lote 4", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 103, "score_global": 7.2, "score_by_pillar": { "Contenido": 7.41, "Agentes": 7.02, "Coding": 7.26, "Razonamiento": 7.16 }, "score_by_suite": { "content_generation": 7.94, "tool_calling": 7.45, "task_management": 7.58, "code_generation": 7.51, "reasoning": 7.23, "summarization": 7.2, "presentation": 7.48, "startup_content": 7.48, "deep_reasoning": 6.91, "customer_support": 7.52, "structured_output": 7.55, "hallucination": 7.6, "creativity": 7.46, "string_precision": 7.48, "news_seo_writing": 6.7, "ocr_extraction": 6.56, "orchestration": 6.11, "multi_turn": 6.65, "policy_adherence": 6.96, "agent_capabilities": 7.2, "strategy": 7.15, "sales_outreach": 7.36, "translation": 7.87, "agent_long_horizon": 7.11 }, "quality_avg": 8.11, "cost_score_avg": 6.73, "speed_score_avg": 6.91, "latency_score_avg": 2.54, "tool_calling_score_avg": 7.11, "judge_score_avg": 4.14, "tokens_per_second": 52.5, "latency_avg_s": 24.79, "total_input_tokens": 250769, "total_output_tokens": 137859 }, { "key": "gpt-4.1-mini", "id": "gpt-4.1-mini", "name": "GPT-4.1 Mini", "tier": "cheap", "provider": "openai_direct", "open_source": false, "license": "", "cost_input_per_M": 0.4, "cost_output_per_M": 1.6, "cost_per_1k_calls_usd": 2.52, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 155, "score_global": 7.17, "score_by_pillar": { "Contenido": 7.34, "Agentes": 6.43, "Coding": 7.72, "Razonamiento": 7.43 }, "score_by_suite": { "content_generation": 7.49, "tool_calling": 6.73, "task_management": 7.36, "code_generation": 7.53, "reasoning": 7.61, "summarization": 6.88, "presentation": 7.23, "startup_content": 7.7, "deep_reasoning": 7.34, "customer_support": 5.46, "structured_output": 7.49, "hallucination": 7.33, "creativity": 7.17, "string_precision": 8.28, "news_seo_writing": 6.92, "agent_capabilities": 5.93, "ocr_extraction": 7.02, "orchestration": 6.45, "multi_turn": 6.88, "policy_adherence": 7.16, "strategy": 7.6, "sales_outreach": 7.87, "translation": 7.87 }, "quality_avg": 7.49, "cost_score_avg": 7.83, "speed_score_avg": 7.65, "latency_score_avg": 3.52, "tool_calling_score_avg": 6.98, "judge_score_avg": 4.18, "tokens_per_second": 71.1, "latency_avg_s": 7.53, "total_input_tokens": 35337, "total_output_tokens": 80931 }, { "key": "hermes-4-405b", "id": "nousresearch/hermes-4-405b", "name": "Hermes 4 405B", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Llama 3 community", "cost_input_per_M": 1.0, "cost_output_per_M": 3.0, "cost_per_1k_calls_usd": 4.8, "subscriptions": [], "notes": "Flagship Hermes 4 con reasoning híbrido. Sin reasoning explícito en este config.", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 74, "score_global": 7.17, "score_by_pillar": { "Contenido": 7.19, "Agentes": 7.11, "Coding": 7.2, "Razonamiento": 7.13 }, "score_by_suite": { "content_generation": 7.31, "task_management": 7.48, "code_generation": 7.18, "reasoning": 7.39, "summarization": 6.56, "presentation": 7.17, "startup_content": 7.22, "deep_reasoning": 6.86, "structured_output": 7.55, "hallucination": 7.31, "creativity": 7.46, "string_precision": 7.41, "news_seo_writing": 6.47, "ocr_extraction": 6.68, "multi_turn": 7.14, "policy_adherence": 6.71, "agent_capabilities": 7.47, "strategy": 7.22, "sales_outreach": 7.69, "translation": 7.8 }, "quality_avg": 8.05, "cost_score_avg": 7.62, "speed_score_avg": 5.65, "latency_score_avg": 2.04, "tool_calling_score_avg": 7.0, "judge_score_avg": 4.18, "tokens_per_second": 33.5, "latency_avg_s": 15.37, "total_input_tokens": 16394, "total_output_tokens": 39971 }, { "key": "nim-devstral-2-123b", "id": "mistralai/devstral-2-123b-instruct-2512", "name": "Devstral 2 123B (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Mismo modelo que probamos via OpenRouter - comparar provider stability", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 68, "score_global": 7.16, "score_by_pillar": { "Contenido": 7.26, "Agentes": 6.91, "Coding": 7.22, "Razonamiento": 7.22 }, "score_by_suite": { "content_generation": 7.56, "tool_calling": 6.58, "task_management": 7.61, "code_generation": 7.38, "reasoning": 7.4, "presentation": 7.18, "startup_content": 7.47, "deep_reasoning": 7.05, "customer_support": 6.68, "structured_output": 7.49, "hallucination": 7.38, "creativity": 7.58, "string_precision": 7.48, "news_seo_writing": 6.57, "ocr_extraction": 6.59, "orchestration": 6.77, "multi_turn": 7.3 }, "quality_avg": 7.98, "cost_score_avg": 7.5, "speed_score_avg": 6.22, "latency_score_avg": 2.56, "tool_calling_score_avg": 6.87, "judge_score_avg": 4.1, "tokens_per_second": 42.3, "latency_avg_s": 17.58, "total_input_tokens": 19474, "total_output_tokens": 45580 }, { "key": "llama-4-maverick", "id": "meta-llama/llama-4-maverick", "name": "Llama 4 Maverick", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Llama Community", "cost_input_per_M": 0.5, "cost_output_per_M": 1.0, "cost_per_1k_calls_usd": 1.65, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 134, "score_global": 7.13, "score_by_pillar": { "Contenido": 7.13, "Agentes": 6.59, "Coding": 7.51, "Razonamiento": 7.14 }, "score_by_suite": { "content_generation": 7.08, "tool_calling": 5.73, "task_management": 7.26, "code_generation": 7.25, "reasoning": 7.23, "summarization": 6.95, "presentation": 7.2, "startup_content": 7.5, "deep_reasoning": 7.01, "customer_support": 5.12, "structured_output": 7.24, "hallucination": 7.34, "creativity": 6.89, "string_precision": 8.09, "news_seo_writing": 6.66, "agent_capabilities": 7.47, "ocr_extraction": 6.98, "multi_turn": 6.77, "policy_adherence": 7.26, "strategy": 7.1, "sales_outreach": 7.69, "translation": 7.75 }, "quality_avg": 7.93, "cost_score_avg": 7.58, "speed_score_avg": 6.01, "latency_score_avg": 2.42, "tool_calling_score_avg": 6.81, "judge_score_avg": 4.16, "tokens_per_second": 47.1, "latency_avg_s": 13.17, "total_input_tokens": 31998, "total_output_tokens": 72629 }, { "key": "devstral-2", "id": "mistralai/devstral-2512", "name": "Devstral 2 (Dic 2025)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.4, "cost_output_per_M": 2.0, "cost_per_1k_calls_usd": 3.12, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 160, "score_global": 7.09, "score_by_pillar": { "Contenido": 7.53, "Agentes": 6.86, "Coding": 7.6, "Razonamiento": 7.37 }, "score_by_suite": { "content_generation": 7.84, "tool_calling": 6.45, "task_management": 7.87, "code_generation": 7.88, "reasoning": 7.59, "summarization": 7.42, "presentation": 7.33, "startup_content": 7.61, "deep_reasoning": 7.06, "customer_support": 5.07, "structured_output": 7.79, "hallucination": 7.55, "creativity": 7.69, "string_precision": 7.81, "news_seo_writing": 6.7, "ocr_extraction": 6.97, "orchestration": 6.69, "multi_turn": 7.57, "policy_adherence": 7.42, "agent_capabilities": 7.18, "strategy": 7.6, "sales_outreach": 7.92, "translation": 7.98, "agent_long_horizon": 7.72, "niah_es": 6.75, "niah_es_lite": 6.59 }, "quality_avg": 7.8, "cost_score_avg": 6.89, "speed_score_avg": 6.28, "latency_score_avg": 3.83, "tool_calling_score_avg": 6.98, "judge_score_avg": 4.22, "tokens_per_second": 53.0, "latency_avg_s": 13.15, "total_input_tokens": 1757050, "total_output_tokens": 138276 }, { "key": "claude-opus-4.6", "id": "anthropic/claude-opus-4-6", "name": "Claude Opus 4.6", "tier": "premium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 15.0, "cost_output_per_M": 75.0, "cost_per_1k_calls_usd": 117.0, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 246, "score_global": 7.09, "score_by_pillar": { "Contenido": 7.07, "Agentes": 6.91, "Coding": 7.2, "Razonamiento": 7.33 }, "score_by_suite": { "content_generation": 7.02, "tool_calling": 6.38, "task_management": 7.56, "code_generation": 7.42, "reasoning": 7.43, "summarization": 7.23, "presentation": 7.22, "startup_content": 7.24, "deep_reasoning": 7.22, "customer_support": 7.27, "structured_output": 7.1, "hallucination": 7.56, "creativity": 6.98, "string_precision": 7.35, "news_seo_writing": 6.6, "agent_capabilities": 6.86, "ocr_extraction": 6.8, "orchestration": 6.48, "multi_turn": 6.86, "policy_adherence": 7.08, "strategy": 7.14, "sales_outreach": 7.49, "translation": 7.33 }, "quality_avg": 8.04, "cost_score_avg": 6.83, "speed_score_avg": 6.51, "latency_score_avg": 1.8, "tool_calling_score_avg": 7.14, "judge_score_avg": 4.18, "tokens_per_second": 46.7, "latency_avg_s": 19.14, "total_input_tokens": 105497, "total_output_tokens": 232473 }, { "key": "mistral-large", "id": "mistralai/mistral-large", "name": "Mistral Large", "tier": "medium", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 2.0, "cost_output_per_M": 6.0, "cost_per_1k_calls_usd": 9.6, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 152, "score_global": 7.08, "score_by_pillar": { "Contenido": 7.14, "Agentes": 6.35, "Coding": 7.74, "Razonamiento": 7.38 }, "score_by_suite": { "content_generation": 7.21, "tool_calling": 6.45, "task_management": 7.56, "code_generation": 7.87, "reasoning": 7.55, "summarization": 7.28, "presentation": 7.41, "startup_content": 7.24, "deep_reasoning": 7.22, "customer_support": 4.29, "structured_output": 7.43, "hallucination": 7.56, "creativity": 6.83, "string_precision": 8.19, "news_seo_writing": 6.55, "agent_capabilities": 6.58, "ocr_extraction": 6.97, "orchestration": 6.75, "multi_turn": 6.75, "policy_adherence": 6.96, "strategy": 7.35, "sales_outreach": 7.63, "translation": 7.83 }, "quality_avg": 7.81, "cost_score_avg": 7.11, "speed_score_avg": 6.81, "latency_score_avg": 2.36, "tool_calling_score_avg": 7.01, "judge_score_avg": 4.21, "tokens_per_second": 49.8, "latency_avg_s": 15.92, "total_input_tokens": 44800, "total_output_tokens": 128547 }, { "key": "gemma-4-26b", "id": "google/gemma-4-26b-a4b-it", "name": "Gemma 4 26B MoE (3.8B activos)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.15, "cost_output_per_M": 0.3, "cost_per_1k_calls_usd": 0.495, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 104, "score_global": 7.06, "score_by_pillar": { "Contenido": 7.25, "Agentes": 6.62, "Coding": 7.31, "Razonamiento": 7.24 }, "score_by_suite": { "content_generation": 7.23, "tool_calling": 6.11, "startup_content": 7.31, "task_management": 7.58, "code_generation": 7.56, "reasoning": 7.42, "summarization": 6.75, "presentation": 7.21, "deep_reasoning": 7.06, "customer_support": 5.68, "structured_output": 7.53, "hallucination": 7.37, "creativity": 7.54, "string_precision": 7.47, "news_seo_writing": 6.9, "ocr_extraction": 6.75, "orchestration": 7.07, "multi_turn": 7.11, "policy_adherence": 7.08, "agent_capabilities": 6.38, "strategy": 7.29, "sales_outreach": 7.71, "translation": 7.24 }, "quality_avg": 7.8, "cost_score_avg": 7.6, "speed_score_avg": 5.91, "latency_score_avg": 2.18, "tool_calling_score_avg": 7.2, "judge_score_avg": 4.2, "tokens_per_second": 44.3, "latency_avg_s": 17.38, "total_input_tokens": 31577, "total_output_tokens": 68689 }, { "key": "nim-nemotron-3-nano-omni-reasoning", "id": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning", "name": "Nemotron 3 Nano Omni 30B-A3B Reasoning (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "NVIDIA Open License", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Lanzado 20 abril 2026. MoE 30B totales / 3B activos (A3B). Multimodal (texto+imagen+audio+video) + reasoning. Comparar contra Nano 30B normal y Nano 9B v2 — sirve para medir el costo de razonar y la ganancia de multimodal en single-turn texto.", "tested": true, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 91, "score_global": 7.04, "score_by_pillar": { "Contenido": 7.23, "Agentes": 6.67, "Coding": 7.24, "Razonamiento": 7.14 }, "score_by_suite": { "content_generation": 7.51, "tool_calling": 6.83, "task_management": 6.97, "code_generation": 7.25, "reasoning": 7.43, "summarization": 7.12, "presentation": 7.18, "startup_content": 7.62, "deep_reasoning": 6.8, "customer_support": 6.69, "structured_output": 7.55, "hallucination": 7.49, "creativity": 7.16, "string_precision": 7.66, "news_seo_writing": 6.69, "ocr_extraction": 6.48, "orchestration": 6.16, "multi_turn": 6.88, "policy_adherence": 6.95, "agent_capabilities": 6.46, "strategy": 7.19, "sales_outreach": 7.23, "translation": 7.3 }, "quality_avg": 7.75, "cost_score_avg": 5.81, "speed_score_avg": 9.89, "latency_score_avg": 2.57, "tool_calling_score_avg": 6.87, "judge_score_avg": 4.09, "tokens_per_second": 202.7, "latency_avg_s": 10.17, "total_input_tokens": 35902, "total_output_tokens": 198864 }, { "key": "mistral-nemo", "id": "mistralai/mistral-nemo", "name": "Mistral Nemo", "tier": "ultra_cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.02, "cost_output_per_M": 0.02, "cost_per_1k_calls_usd": 0.036, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 122, "score_global": 7.03, "score_by_pillar": { "Razonamiento": 7.09, "Agentes": 6.38, "Coding": 7.44, "Contenido": 7.24 }, "score_by_suite": { "deep_reasoning": 6.74, "customer_support": 4.98, "structured_output": 7.28, "hallucination": 7.38, "creativity": 6.93, "string_precision": 7.91, "news_seo_writing": 7.04, "content_generation": 7.79, "tool_calling": 7.06, "task_management": 7.68, "code_generation": 7.01, "reasoning": 7.31, "summarization": 6.76, "presentation": 7.59, "startup_content": 7.44, "ocr_extraction": 6.95, "orchestration": 6.36, "multi_turn": 6.76, "policy_adherence": 7.13, "agent_capabilities": 6.41, "strategy": 7.73, "sales_outreach": 7.38, "translation": 7.58 }, "quality_avg": 6.97, "cost_score_avg": 9.0, "speed_score_avg": 4.64, "latency_score_avg": 2.66, "tool_calling_score_avg": 6.67, "judge_score_avg": 3.91, "tokens_per_second": 33.7, "latency_avg_s": 17.47, "total_input_tokens": 34975, "total_output_tokens": 42864 }, { "key": "hermes-4-70b", "id": "nousresearch/hermes-4-70b", "name": "Hermes 4 70B", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Llama 3 community", "cost_input_per_M": 0.13, "cost_output_per_M": 0.4, "cost_per_1k_calls_usd": 0.639, "subscriptions": [], "notes": "Hybrid reasoning mode. Open-source de Nous Research. Sin reasoning explícito en este config.", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 131, "score_global": 6.99, "score_by_pillar": { "Contenido": 7.35, "Agentes": 7.2, "Coding": 7.46, "Razonamiento": 7.26 }, "score_by_suite": { "content_generation": 7.77, "task_management": 7.52, "code_generation": 7.29, "reasoning": 7.42, "summarization": 7.22, "presentation": 7.24, "startup_content": 7.45, "deep_reasoning": 7.04, "structured_output": 7.9, "hallucination": 7.38, "creativity": 7.44, "string_precision": 7.76, "news_seo_writing": 6.64, "ocr_extraction": 6.89, "multi_turn": 6.92, "policy_adherence": 7.18, "agent_capabilities": 7.42, "strategy": 7.43, "sales_outreach": 7.64, "translation": 7.61, "agent_long_horizon": 7.37, "niah_es_lite": 6.3 }, "quality_avg": 7.84, "cost_score_avg": 6.16, "speed_score_avg": 6.3, "latency_score_avg": 3.82, "tool_calling_score_avg": 7.0, "judge_score_avg": 4.12, "tokens_per_second": 50.0, "latency_avg_s": 7.32, "total_input_tokens": 1492840, "total_output_tokens": 60488 }, { "key": "kimi-k2", "id": "moonshotai/kimi-k2", "name": "Kimi K2", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Modified MIT", "cost_input_per_M": 0.2, "cost_output_per_M": 0.8, "cost_per_1k_calls_usd": 1.26, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 138, "score_global": 6.96, "score_by_pillar": { "Contenido": 6.86, "Agentes": 6.63, "Coding": 7.34, "Razonamiento": 7.1 }, "score_by_suite": { "content_generation": 6.75, "tool_calling": 6.2, "task_management": 6.89, "code_generation": 6.99, "reasoning": 7.52, "summarization": 6.69, "presentation": 6.75, "startup_content": 7.31, "deep_reasoning": 6.99, "customer_support": 6.93, "structured_output": 7.16, "hallucination": 6.99, "creativity": 6.36, "string_precision": 7.96, "news_seo_writing": 6.71, "agent_capabilities": 6.29, "ocr_extraction": 6.57, "orchestration": 6.91, "multi_turn": 6.94, "policy_adherence": 7.1, "strategy": 6.84, "sales_outreach": 7.37, "translation": 7.53 }, "quality_avg": 7.76, "cost_score_avg": 7.36, "speed_score_avg": 4.9, "latency_score_avg": 1.99, "tool_calling_score_avg": 7.21, "judge_score_avg": 4.12, "tokens_per_second": 28.3, "latency_avg_s": 20.64, "total_input_tokens": 44693, "total_output_tokens": 88841 }, { "key": "qwen3-coder", "id": "qwen/qwen3-coder", "name": "Qwen3 Coder", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.2, "cost_output_per_M": 0.6, "cost_per_1k_calls_usd": 0.96, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 155, "score_global": 6.94, "score_by_pillar": { "Contenido": 7.08, "Agentes": 6.0, "Coding": 7.64, "Razonamiento": 7.37 }, "score_by_suite": { "content_generation": 6.99, "tool_calling": 6.25, "task_management": 7.4, "code_generation": 7.8, "reasoning": 7.43, "summarization": 6.72, "presentation": 7.47, "startup_content": 7.52, "deep_reasoning": 7.35, "customer_support": 4.12, "structured_output": 7.23, "hallucination": 7.34, "creativity": 6.74, "string_precision": 8.13, "news_seo_writing": 6.71, "agent_capabilities": 5.56, "ocr_extraction": 6.89, "orchestration": 6.35, "multi_turn": 6.84, "policy_adherence": 6.98, "strategy": 7.35, "sales_outreach": 7.39, "translation": 7.63 }, "quality_avg": 7.57, "cost_score_avg": 7.46, "speed_score_avg": 5.97, "latency_score_avg": 2.54, "tool_calling_score_avg": 6.89, "judge_score_avg": 4.22, "tokens_per_second": 53.8, "latency_avg_s": 17.65, "total_input_tokens": 57007, "total_output_tokens": 95691 }, { "key": "mimo-v2-omni", "id": "xiaomi/mimo-v2-omni", "name": "MiMo-V2-Omni (multimodal)", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.4, "cost_output_per_M": 2.0, "cost_per_1k_calls_usd": 3.12, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 103, "score_global": 6.94, "score_by_pillar": { "Contenido": 7.29, "Agentes": 6.63, "Coding": 6.77, "Razonamiento": 6.86 }, "score_by_suite": { "content_generation": 7.88, "tool_calling": 6.86, "task_management": 7.9, "code_generation": 6.43, "reasoning": 7.54, "summarization": 6.76, "presentation": 7.51, "startup_content": 7.6, "deep_reasoning": 5.98, "customer_support": 6.2, "structured_output": 6.79, "hallucination": 7.35, "creativity": 7.51, "string_precision": 7.5, "news_seo_writing": 6.49, "ocr_extraction": 6.14, "orchestration": 5.48, "multi_turn": 6.88, "policy_adherence": 7.24, "agent_capabilities": 6.52, "strategy": 7.45, "sales_outreach": 7.35, "translation": 7.11, "agent_long_horizon": 7.27 }, "quality_avg": 7.52, "cost_score_avg": 7.0, "speed_score_avg": 8.41, "latency_score_avg": 1.84, "tool_calling_score_avg": 6.91, "judge_score_avg": 4.0, "tokens_per_second": 85.8, "latency_avg_s": 18.52, "total_input_tokens": 208736, "total_output_tokens": 173093 }, { "key": "grok-4.20", "id": "x-ai/grok-4.20", "name": "Grok 4.20", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 2.0, "cost_output_per_M": 6.0, "cost_per_1k_calls_usd": 9.6, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 91, "score_global": 6.91, "score_by_pillar": { "Contenido": 7.13, "Agentes": 6.48, "Coding": 7.18, "Razonamiento": 7.01 }, "score_by_suite": { "content_generation": 7.36, "tool_calling": 6.97, "task_management": 7.51, "code_generation": 7.59, "reasoning": 7.43, "summarization": 7.12, "presentation": 7.11, "startup_content": 7.43, "deep_reasoning": 6.76, "customer_support": 6.01, "structured_output": 7.19, "hallucination": 7.0, "creativity": 7.23, "string_precision": 7.18, "news_seo_writing": 6.41, "ocr_extraction": 6.85, "orchestration": 5.6, "multi_turn": 6.76, "policy_adherence": 6.84, "agent_capabilities": 6.2, "strategy": 7.12, "sales_outreach": 7.37, "translation": 7.2 }, "quality_avg": 7.83, "cost_score_avg": 6.32, "speed_score_avg": 6.26, "latency_score_avg": 2.46, "tool_calling_score_avg": 6.96, "judge_score_avg": 4.1, "tokens_per_second": 50.8, "latency_avg_s": 10.82, "total_input_tokens": 37549, "total_output_tokens": 53178 }, { "key": "nim-qwen3.5-397b", "id": "qwen/qwen3.5-397b-a17b", "name": "Qwen 3.5 397B (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Mismo modelo que Cristian usa via Ollama Cloud para producción — comparar", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 97, "score_global": 6.9, "score_by_pillar": { "Contenido": 6.95, "Agentes": 7.03, "Coding": 6.89, "Razonamiento": 6.94 }, "score_by_suite": { "content_generation": 7.05, "tool_calling": 6.92, "task_management": 7.28, "code_generation": 7.18, "reasoning": 7.12, "summarization": 6.29, "presentation": 6.99, "startup_content": 7.15, "deep_reasoning": 6.83, "customer_support": 7.44, "structured_output": 7.14, "hallucination": 7.1, "creativity": 7.33, "string_precision": 6.89, "news_seo_writing": 6.36, "ocr_extraction": 6.47, "orchestration": 6.5, "multi_turn": 6.66, "policy_adherence": 7.05, "agent_capabilities": 7.44, "strategy": 6.84, "sales_outreach": 7.06, "translation": 7.26, "agent_long_horizon": 5.96 }, "quality_avg": 8.07, "cost_score_avg": 6.81, "speed_score_avg": 3.71, "latency_score_avg": 1.47, "tool_calling_score_avg": 7.22, "judge_score_avg": 4.18, "tokens_per_second": 20.7, "latency_avg_s": 57.29, "total_input_tokens": 166523, "total_output_tokens": 110548 }, { "key": "nim-nemotron-nano-9b-v2", "id": "nvidia/nvidia-nemotron-nano-9b-v2", "name": "Nemotron Nano 9B v2 (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": false, "license": "NVIDIA Open License", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Variante chica de Nemotron 3, comparar con Nano 30B", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 91, "score_global": 6.86, "score_by_pillar": { "Contenido": 7.16, "Agentes": 6.44, "Coding": 7.05, "Razonamiento": 6.86 }, "score_by_suite": { "content_generation": 7.48, "tool_calling": 6.68, "task_management": 7.42, "code_generation": 7.35, "reasoning": 7.42, "summarization": 6.75, "presentation": 7.37, "startup_content": 7.42, "deep_reasoning": 6.17, "customer_support": 6.75, "structured_output": 7.29, "hallucination": 7.3, "creativity": 7.19, "string_precision": 7.23, "news_seo_writing": 6.48, "ocr_extraction": 6.4, "orchestration": 4.92, "multi_turn": 6.66, "policy_adherence": 6.96, "agent_capabilities": 6.37, "strategy": 7.23, "sales_outreach": 7.34, "translation": 7.29 }, "quality_avg": 7.73, "cost_score_avg": 6.25, "speed_score_avg": 7.56, "latency_score_avg": 1.27, "tool_calling_score_avg": 7.1, "judge_score_avg": 4.1, "tokens_per_second": 54.3, "latency_avg_s": 26.4, "total_input_tokens": 31101, "total_output_tokens": 132905 }, { "key": "gpt-5.4", "id": "gpt-5.4", "name": "GPT-5.4", "tier": "premium", "provider": "openai_direct", "open_source": false, "license": "", "cost_input_per_M": 5.0, "cost_output_per_M": 15.0, "cost_per_1k_calls_usd": 24.0, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 150, "score_global": 6.85, "score_by_pillar": { "Contenido": 6.77, "Agentes": 6.35, "Coding": 7.58, "Razonamiento": 6.87 }, "score_by_suite": { "content_generation": 7.17, "tool_calling": 6.68, "task_management": 6.73, "code_generation": 7.25, "reasoning": 5.96, "summarization": 6.88, "presentation": 7.29, "startup_content": 5.97, "deep_reasoning": 6.98, "customer_support": 5.21, "structured_output": 7.29, "hallucination": 7.36, "creativity": 6.27, "string_precision": 8.25, "news_seo_writing": 6.78, "ocr_extraction": 6.93, "orchestration": 6.92, "multi_turn": 6.79, "policy_adherence": 7.11, "agent_capabilities": 5.65, "strategy": 7.29, "sales_outreach": 7.84, "translation": 7.72 }, "quality_avg": 7.09, "cost_score_avg": 7.35, "speed_score_avg": 7.17, "latency_score_avg": 2.84, "tool_calling_score_avg": 7.14, "judge_score_avg": 4.18, "tokens_per_second": 56.8, "latency_avg_s": 12.35, "total_input_tokens": 34875, "total_output_tokens": 116069 }, { "key": "nim-ministral-14b", "id": "mistralai/ministral-14b-instruct-2512", "name": "Ministral 14B (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Mistral chico nuevo dic 2025 - comparar con Mistral Small 4", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 90, "score_global": 6.84, "score_by_pillar": { "Contenido": 7.07, "Agentes": 6.6, "Coding": 6.92, "Razonamiento": 6.77 }, "score_by_suite": { "content_generation": 7.33, "tool_calling": 7.13, "task_management": 7.26, "code_generation": 6.98, "reasoning": 7.15, "summarization": 7.13, "presentation": 6.96, "startup_content": 7.2, "deep_reasoning": 6.48, "customer_support": 6.05, "structured_output": 6.59, "hallucination": 6.87, "creativity": 7.18, "string_precision": 7.36, "news_seo_writing": 6.33, "ocr_extraction": 6.61, "orchestration": 6.39, "multi_turn": 6.72, "policy_adherence": 6.78, "agent_capabilities": 6.16, "strategy": 6.87, "sales_outreach": 7.14, "translation": 7.55 }, "quality_avg": 8.02, "cost_score_avg": 6.89, "speed_score_avg": 3.9, "latency_score_avg": 1.69, "tool_calling_score_avg": 6.72, "judge_score_avg": 4.12, "tokens_per_second": 21.5, "latency_avg_s": 49.5, "total_input_tokens": 27548, "total_output_tokens": 86481 }, { "key": "nemotron-super", "id": "nvidia/nemotron-3-super-120b-a12b", "name": "Nemotron 3 Super", "tier": "ultra_cheap", "provider": "openrouter", "open_source": true, "license": "NVIDIA Open", "cost_input_per_M": 0.1, "cost_output_per_M": 0.5, "cost_per_1k_calls_usd": 0.78, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 91, "score_global": 6.83, "score_by_pillar": { "Contenido": 7.45, "Agentes": 5.75, "Coding": 7.29, "Razonamiento": 7.19 }, "score_by_suite": { "content_generation": 7.66, "tool_calling": 6.06, "task_management": 7.45, "code_generation": 7.65, "reasoning": 7.57, "summarization": 7.1, "presentation": 7.89, "startup_content": 7.65, "deep_reasoning": 6.8, "customer_support": 4.36, "structured_output": 7.48, "hallucination": 7.37, "creativity": 7.5, "string_precision": 7.37, "news_seo_writing": 6.7, "ocr_extraction": 6.75, "orchestration": 4.49, "multi_turn": 7.32, "policy_adherence": 7.21, "agent_capabilities": 4.43, "strategy": 7.42, "sales_outreach": 7.72, "translation": 7.72 }, "quality_avg": 7.17, "cost_score_avg": 8.38, "speed_score_avg": 4.42, "latency_score_avg": 1.26, "tool_calling_score_avg": 6.85, "judge_score_avg": 4.27, "tokens_per_second": 30.6, "latency_avg_s": 46.1, "total_input_tokens": 35068, "total_output_tokens": 96589 }, { "key": "nim-mistral-large-3", "id": "mistralai/mistral-large-3-675b-instruct-2512", "name": "Mistral Large 3 675B (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Flagship Mistral diciembre 2025. Comparar contra Mistral Small 4 (#4 ranking).", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 87, "score_global": 6.83, "score_by_pillar": { "Contenido": 6.95, "Agentes": 6.67, "Coding": 6.9, "Razonamiento": 6.84 }, "score_by_suite": { "content_generation": 7.44, "tool_calling": 6.92, "task_management": 7.09, "code_generation": 7.16, "reasoning": 6.97, "summarization": 6.97, "presentation": 6.9, "startup_content": 7.0, "deep_reasoning": 6.61, "customer_support": 6.44, "structured_output": 6.88, "hallucination": 6.99, "creativity": 6.97, "string_precision": 7.07, "news_seo_writing": 5.85, "ocr_extraction": 6.51, "orchestration": 6.91, "multi_turn": 6.46, "policy_adherence": 6.63, "agent_capabilities": 6.43, "strategy": 7.03, "sales_outreach": 7.33, "translation": 7.41 }, "quality_avg": 8.18, "cost_score_avg": 7.25, "speed_score_avg": 2.14, "latency_score_avg": 1.31, "tool_calling_score_avg": 6.88, "judge_score_avg": 4.2, "tokens_per_second": 10.6, "latency_avg_s": 141.03, "total_input_tokens": 27281, "total_output_tokens": 67680 }, { "key": "mimo-v2-pro", "id": "xiaomi/mimo-v2-pro", "name": "MiMo-V2-Pro", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "Proprietary", "cost_input_per_M": 1.0, "cost_output_per_M": 3.0, "cost_per_1k_calls_usd": 4.8, "subscriptions": [], "notes": "API-only Xiaomi (NO confundir con MiMo V2 Flash que sí es MIT en HF). El Pro no se publica en HuggingFace.", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 91, "score_global": 6.8, "score_by_pillar": { "Contenido": 7.21, "Agentes": 6.6, "Coding": 6.65, "Razonamiento": 6.61 }, "score_by_suite": { "content_generation": 7.68, "tool_calling": 6.2, "task_management": 7.6, "code_generation": 6.87, "reasoning": 7.18, "summarization": 7.11, "presentation": 7.41, "startup_content": 7.44, "deep_reasoning": 5.58, "customer_support": 6.96, "structured_output": 7.06, "hallucination": 7.34, "creativity": 7.3, "string_precision": 7.36, "news_seo_writing": 6.36, "ocr_extraction": 5.29, "orchestration": 6.14, "multi_turn": 6.66, "policy_adherence": 7.0, "agent_capabilities": 6.1, "strategy": 7.35, "sales_outreach": 7.49, "translation": 7.18 }, "quality_avg": 7.52, "cost_score_avg": 6.45, "speed_score_avg": 6.92, "latency_score_avg": 1.47, "tool_calling_score_avg": 7.23, "judge_score_avg": 4.15, "tokens_per_second": 51.5, "latency_avg_s": 20.89, "total_input_tokens": 43909, "total_output_tokens": 97394 }, { "key": "local-nemotron3-base-33b", "id": "nemotron3:33b-q4_K_M", "name": "Nemotron 3 Base 33B (DGX Spark Q4_K_M)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "NVIDIA Open License", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Nemotron 3 base 33B (NO la versión Omni Reasoning) en Q4_K_M via Ollama oficial. Comparar con Nemotron 3 Super 120B (DGX, también Q4) y Nemotron 3 Nano Omni Reasoning cuando esté disponible.", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 103, "score_global": 6.77, "score_by_pillar": { "Contenido": 7.09, "Agentes": 6.34, "Coding": 6.93, "Razonamiento": 6.93 }, "score_by_suite": { "content_generation": 7.42, "tool_calling": 6.11, "task_management": 6.98, "code_generation": 7.18, "reasoning": 7.33, "summarization": 6.93, "presentation": 7.14, "startup_content": 7.4, "deep_reasoning": 6.64, "customer_support": 6.37, "structured_output": 7.41, "hallucination": 7.11, "creativity": 6.95, "string_precision": 7.04, "news_seo_writing": 6.47, "ocr_extraction": 6.21, "orchestration": 6.0, "multi_turn": 6.49, "policy_adherence": 6.78, "agent_capabilities": 5.98, "strategy": 6.95, "sales_outreach": 7.38, "translation": 7.1, "agent_long_horizon": 6.65 }, "quality_avg": 7.83, "cost_score_avg": 5.27, "speed_score_avg": 7.92, "latency_score_avg": 1.52, "tool_calling_score_avg": 6.74, "judge_score_avg": 4.07, "tokens_per_second": 62.7, "latency_avg_s": 58.29, "total_input_tokens": 262662, "total_output_tokens": 344974 }, { "key": "gpt-4.1", "id": "gpt-4.1", "name": "GPT-4.1", "tier": "medium", "provider": "openai_direct", "open_source": false, "license": "", "cost_input_per_M": 2.0, "cost_output_per_M": 8.0, "cost_per_1k_calls_usd": 12.6, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 254, "score_global": 6.74, "score_by_pillar": { "Contenido": 7.43, "Agentes": 6.6, "Coding": 7.92, "Razonamiento": 7.52 }, "score_by_suite": { "content_generation": 7.55, "tool_calling": 6.83, "task_management": 7.59, "code_generation": 7.87, "reasoning": 7.73, "summarization": 7.35, "presentation": 7.52, "startup_content": 7.65, "deep_reasoning": 7.4, "customer_support": 5.6, "structured_output": 7.65, "hallucination": 7.56, "creativity": 7.11, "string_precision": 8.36, "news_seo_writing": 7.03, "agent_capabilities": 5.81, "ocr_extraction": 7.37, "orchestration": 6.93, "multi_turn": 7.47, "policy_adherence": 7.29, "strategy": 7.52, "sales_outreach": 7.95, "translation": 7.95, "agent_long_horizon": 7.13, "niah_es": 5.84, "niah_es_1m": 4.91 }, "quality_avg": 7.62, "cost_score_avg": 5.66, "speed_score_avg": 6.24, "latency_score_avg": 3.72, "tool_calling_score_avg": 7.06, "judge_score_avg": 4.28, "tokens_per_second": 60.8, "latency_avg_s": 21.07, "total_input_tokens": 19675632, "total_output_tokens": 113741 }, { "key": "nim-glm5", "id": "z-ai/glm5", "name": "GLM 5 (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Z.AI flagship base, comparar con 5.1", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 90, "score_global": 6.73, "score_by_pillar": { "Contenido": 6.84, "Agentes": 6.54, "Coding": 6.86, "Razonamiento": 6.74 }, "score_by_suite": { "content_generation": 7.06, "tool_calling": 6.0, "task_management": 7.27, "code_generation": 7.01, "reasoning": 7.13, "summarization": 6.36, "presentation": 6.83, "startup_content": 7.0, "deep_reasoning": 6.43, "customer_support": 6.96, "structured_output": 7.18, "hallucination": 6.94, "creativity": 6.99, "string_precision": 6.96, "news_seo_writing": 6.09, "ocr_extraction": 6.34, "orchestration": 6.39, "multi_turn": 6.58, "policy_adherence": 6.65, "agent_capabilities": 6.24, "strategy": 6.79, "sales_outreach": 7.24, "translation": 6.98 }, "quality_avg": 7.97, "cost_score_avg": 7.27, "speed_score_avg": 1.72, "latency_score_avg": 1.1, "tool_calling_score_avg": 7.18, "judge_score_avg": 4.09, "tokens_per_second": 7.0, "latency_avg_s": 123.21, "total_input_tokens": 27716, "total_output_tokens": 67035 }, { "key": "glm-5.1", "id": "z-ai/glm-5.1", "name": "GLM-5.1", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 0.95, "cost_output_per_M": 3.15, "cost_per_1k_calls_usd": 5.01, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 182, "score_global": 6.69, "score_by_pillar": { "Contenido": 6.76, "Agentes": 6.56, "Coding": 6.83, "Razonamiento": 6.65 }, "score_by_suite": { "content_generation": 7.15, "tool_calling": 6.23, "task_management": 7.26, "code_generation": 6.96, "reasoning": 7.14, "summarization": 6.57, "presentation": 6.59, "startup_content": 6.94, "deep_reasoning": 6.21, "customer_support": 6.73, "structured_output": 7.1, "hallucination": 6.79, "creativity": 6.89, "string_precision": 6.91, "news_seo_writing": 6.09, "ocr_extraction": 6.39, "orchestration": 6.33, "multi_turn": 6.45, "policy_adherence": 6.68, "agent_capabilities": 6.47, "strategy": 6.92, "sales_outreach": 6.98, "translation": 6.91 }, "quality_avg": 7.88, "cost_score_avg": 6.4, "speed_score_avg": 3.67, "latency_score_avg": 1.08, "tool_calling_score_avg": 7.14, "judge_score_avg": 4.09, "tokens_per_second": 23.4, "latency_avg_s": 82.48, "total_input_tokens": 56627, "total_output_tokens": 243382 }, { "key": "nim-glm-5.1", "id": "z-ai/glm-5.1", "name": "GLM 5.1 (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Z.AI agentic, no tenemos. Variante GLM 5 con mejoras", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 182, "score_global": 6.69, "score_by_pillar": { "Contenido": 6.76, "Agentes": 6.56, "Coding": 6.83, "Razonamiento": 6.65 }, "score_by_suite": { "content_generation": 7.15, "tool_calling": 6.23, "task_management": 7.26, "code_generation": 6.96, "reasoning": 7.14, "summarization": 6.57, "presentation": 6.59, "startup_content": 6.94, "deep_reasoning": 6.21, "customer_support": 6.73, "structured_output": 7.1, "hallucination": 6.79, "creativity": 6.89, "string_precision": 6.91, "news_seo_writing": 6.09, "ocr_extraction": 6.39, "orchestration": 6.33, "multi_turn": 6.45, "policy_adherence": 6.68, "agent_capabilities": 6.47, "strategy": 6.92, "sales_outreach": 6.98, "translation": 6.91 }, "quality_avg": 7.88, "cost_score_avg": 6.4, "speed_score_avg": 3.67, "latency_score_avg": 1.08, "tool_calling_score_avg": 7.14, "judge_score_avg": 4.09, "tokens_per_second": 23.4, "latency_avg_s": 82.48, "total_input_tokens": 56627, "total_output_tokens": 243382 }, { "key": "local-gemma4-31b", "id": "gemma4:31b", "name": "Gemma 4 31B (DGX Spark Q4_K_M)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Gemma Terms", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Q4_K_M en DGX Spark via Ollama. Comparar con Gemma 4 31B NIM y OpenRouter.", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 89, "score_global": 6.69, "score_by_pillar": { "Contenido": 6.71, "Agentes": 6.66, "Coding": 6.6, "Razonamiento": 6.86 }, "score_by_suite": { "content_generation": 7.11, "tool_calling": 6.61, "task_management": 6.95, "code_generation": 6.87, "reasoning": 7.07, "summarization": 6.31, "presentation": 6.73, "startup_content": 6.98, "deep_reasoning": 6.7, "customer_support": 6.88, "structured_output": 6.83, "hallucination": 6.7, "creativity": 6.88, "string_precision": 6.72, "news_seo_writing": 6.18, "ocr_extraction": 6.05, "orchestration": 6.6, "multi_turn": 6.5, "policy_adherence": 6.41, "agent_capabilities": 6.72, "strategy": 7.07, "sales_outreach": 6.94, "translation": 6.5 }, "quality_avg": 8.22, "cost_score_avg": 6.42, "speed_score_avg": 1.98, "latency_score_avg": 1.01, "tool_calling_score_avg": 7.22, "judge_score_avg": 4.25, "tokens_per_second": 9.3, "latency_avg_s": 153.37, "total_input_tokens": 27238, "total_output_tokens": 117243 }, { "key": "gemma4-31b", "id": "gemma4:31b", "name": "Gemma 4 31B (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 89, "score_global": 6.69, "score_by_pillar": { "Contenido": 6.71, "Agentes": 6.66, "Coding": 6.6, "Razonamiento": 6.86 }, "score_by_suite": { "content_generation": 7.11, "tool_calling": 6.61, "task_management": 6.95, "code_generation": 6.87, "reasoning": 7.07, "summarization": 6.31, "presentation": 6.73, "startup_content": 6.98, "deep_reasoning": 6.7, "customer_support": 6.88, "structured_output": 6.83, "hallucination": 6.7, "creativity": 6.88, "string_precision": 6.72, "news_seo_writing": 6.18, "ocr_extraction": 6.05, "orchestration": 6.6, "multi_turn": 6.5, "policy_adherence": 6.41, "agent_capabilities": 6.72, "strategy": 7.07, "sales_outreach": 6.94, "translation": 6.5 }, "quality_avg": 8.22, "cost_score_avg": 6.42, "speed_score_avg": 1.98, "latency_score_avg": 1.01, "tool_calling_score_avg": 7.22, "judge_score_avg": 4.25, "tokens_per_second": 9.3, "latency_avg_s": 153.37, "total_input_tokens": 27238, "total_output_tokens": 117243 }, { "key": "claude-sonnet-4.6", "id": "anthropic/claude-sonnet-4-6", "name": "Claude Sonnet 4.6 (ultimo Anthropic)", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 3.0, "cost_output_per_M": 15.0, "cost_per_1k_calls_usd": 23.4, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 227, "score_global": 6.65, "score_by_pillar": { "Contenido": 7.15, "Agentes": 6.84, "Coding": 7.37, "Razonamiento": 7.24 }, "score_by_suite": { "content_generation": 7.0, "tool_calling": 6.34, "task_management": 7.45, "code_generation": 7.46, "reasoning": 7.4, "summarization": 7.32, "presentation": 7.23, "startup_content": 7.4, "deep_reasoning": 6.95, "customer_support": 7.18, "structured_output": 7.15, "hallucination": 7.61, "creativity": 6.81, "string_precision": 7.64, "news_seo_writing": 6.8, "agent_capabilities": 6.58, "ocr_extraction": 6.85, "orchestration": 6.72, "multi_turn": 6.63, "policy_adherence": 7.32, "strategy": 7.21, "sales_outreach": 7.76, "translation": 7.48, "niah_es_lite": 4.77 }, "quality_avg": 7.38, "cost_score_avg": 5.92, "speed_score_avg": 6.78, "latency_score_avg": 2.19, "tool_calling_score_avg": 7.27, "judge_score_avg": 4.13, "tokens_per_second": 51.0, "latency_avg_s": 14.58, "total_input_tokens": 1565765, "total_output_tokens": 184300 }, { "key": "local-nemotron-3-super-120b", "id": "nemotron-3-super:120b", "name": "Nemotron 3 Super 120B (DGX Spark Q4_K_M)", "tier": "local", "provider": "openrouter", "open_source": false, "license": "NVIDIA Open License", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Q4_K_M en DGX Spark. Modelo gigante, primer test post cold start tarda ~3-5min.", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 90, "score_global": 6.65, "score_by_pillar": { "Contenido": 6.88, "Agentes": 6.38, "Coding": 6.76, "Razonamiento": 6.65 }, "score_by_suite": { "content_generation": 7.06, "tool_calling": 6.23, "task_management": 6.84, "code_generation": 7.04, "reasoning": 6.97, "summarization": 6.65, "presentation": 7.07, "startup_content": 7.07, "deep_reasoning": 6.56, "customer_support": 6.79, "structured_output": 6.98, "hallucination": 6.58, "creativity": 6.84, "string_precision": 6.98, "news_seo_writing": 6.4, "ocr_extraction": 6.09, "orchestration": 5.8, "multi_turn": 6.61, "policy_adherence": 6.66, "agent_capabilities": 6.05, "strategy": 6.59, "sales_outreach": 7.09, "translation": 6.81 }, "quality_avg": 7.96, "cost_score_avg": 6.0, "speed_score_avg": 3.69, "latency_score_avg": 1.09, "tool_calling_score_avg": 7.07, "judge_score_avg": 4.2, "tokens_per_second": 16.8, "latency_avg_s": 123.69, "total_input_tokens": 34733, "total_output_tokens": 157283 }, { "key": "nim-nemotron-super-1.5", "id": "nvidia/llama-3.3-nemotron-super-49b-v1.5", "name": "Nemotron Super 49B v1.5 (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "NVIDIA Open Model", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Versión iterada del Nemotron Super 120B que ya medimos en Lote 2", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 86, "score_global": 6.64, "score_by_pillar": { "Contenido": 6.96, "Agentes": 6.27, "Coding": 6.58, "Razonamiento": 6.86 }, "score_by_suite": { "content_generation": 7.09, "tool_calling": 6.4, "task_management": 7.09, "code_generation": 6.93, "reasoning": 6.85, "summarization": 6.67, "presentation": 6.98, "startup_content": 7.16, "deep_reasoning": 6.8, "customer_support": 5.87, "structured_output": 6.83, "hallucination": 6.71, "creativity": 7.14, "string_precision": 6.66, "news_seo_writing": 6.31, "ocr_extraction": 6.08, "orchestration": 5.6, "multi_turn": 6.25, "policy_adherence": 6.82, "agent_capabilities": 6.43, "strategy": 7.08, "sales_outreach": 7.05, "translation": 7.42 }, "quality_avg": 7.85, "cost_score_avg": 6.05, "speed_score_avg": 4.37, "latency_score_avg": 1.06, "tool_calling_score_avg": 7.19, "judge_score_avg": 4.09, "tokens_per_second": 25.4, "latency_avg_s": 83.0, "total_input_tokens": 29710, "total_output_tokens": 139881 }, { "key": "nim-deepseek-v4-flash", "id": "deepseek-ai/deepseek-v4-flash", "name": "DeepSeek V4 Flash (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Mismo modelo que probamos via OpenRouter — comparar latencia/calidad NIM vs OR", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 153, "score_global": 6.6, "score_by_pillar": { "Contenido": 7.16, "Agentes": 6.85, "Coding": 7.09, "Razonamiento": 7.12 }, "score_by_suite": { "content_generation": 7.62, "tool_calling": 6.61, "task_management": 7.4, "code_generation": 7.18, "reasoning": 7.33, "summarization": 6.64, "presentation": 7.33, "startup_content": 7.21, "deep_reasoning": 6.98, "customer_support": 6.89, "structured_output": 7.45, "hallucination": 7.23, "creativity": 7.4, "string_precision": 7.21, "news_seo_writing": 6.34, "ocr_extraction": 6.45, "orchestration": 7.16, "multi_turn": 6.72, "policy_adherence": 7.0, "agent_capabilities": 6.54, "strategy": 7.07, "sales_outreach": 7.38, "translation": 7.56, "agent_long_horizon": 6.4, "niah_es": 5.92 }, "quality_avg": 7.9, "cost_score_avg": 5.96, "speed_score_avg": 3.14, "latency_score_avg": 1.87, "tool_calling_score_avg": 7.1, "judge_score_avg": 4.17, "tokens_per_second": 17.2, "latency_avg_s": 82.68, "total_input_tokens": 1787023, "total_output_tokens": 113273 }, { "key": "minimax-m2.7", "id": "minimax/minimax-m2.7", "name": "MiniMax M2.7", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.3, "cost_output_per_M": 1.2, "cost_per_1k_calls_usd": 1.89, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 167, "score_global": 6.59, "score_by_pillar": { "Contenido": 6.68, "Agentes": 6.4, "Coding": 7.1, "Razonamiento": 6.07 }, "score_by_suite": { "content_generation": 6.88, "tool_calling": 6.25, "task_management": 6.55, "code_generation": 6.5, "reasoning": 7.04, "summarization": 6.38, "presentation": 7.13, "startup_content": 7.16, "deep_reasoning": 4.87, "customer_support": 6.47, "structured_output": 7.01, "hallucination": 7.05, "creativity": 5.71, "string_precision": 7.82, "news_seo_writing": 6.62, "agent_capabilities": 6.3, "ocr_extraction": 6.48, "orchestration": 6.01, "multi_turn": 6.76, "policy_adherence": 6.72, "strategy": 6.97, "sales_outreach": 6.81, "translation": 7.06, "agent_long_horizon": 6.66 }, "quality_avg": 7.3, "cost_score_avg": 6.43, "speed_score_avg": 5.75, "latency_score_avg": 1.54, "tool_calling_score_avg": 7.06, "judge_score_avg": 4.11, "tokens_per_second": 39.5, "latency_avg_s": 33.79, "total_input_tokens": 162597, "total_output_tokens": 208835 }, { "key": "qwen-3.6-plus", "id": "qwen/qwen3.6-plus", "name": "Qwen 3.6 Plus", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "Proprietary", "cost_input_per_M": 0.33, "cost_output_per_M": 0.65, "cost_per_1k_calls_usd": 1.074, "subscriptions": [], "notes": "Plus = API-only propietario de Alibaba (NO confundir con Qwen 3.6 base que es Apache 2.0)", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 150, "score_global": 6.56, "score_by_pillar": { "Contenido": 6.64, "Agentes": 5.74, "Coding": 7.1, "Razonamiento": 6.98 }, "score_by_suite": { "content_generation": 6.57, "tool_calling": 6.0, "task_management": 7.03, "code_generation": 7.38, "reasoning": 7.27, "summarization": 6.45, "presentation": 7.05, "startup_content": 7.21, "deep_reasoning": 7.01, "customer_support": 4.75, "structured_output": 6.71, "hallucination": 6.63, "creativity": 6.12, "string_precision": 7.5, "news_seo_writing": 6.27, "ocr_extraction": 6.3, "orchestration": 4.89, "multi_turn": 6.6, "policy_adherence": 6.65, "agent_capabilities": 4.77, "strategy": 6.98, "sales_outreach": 7.28, "translation": 6.68 }, "quality_avg": 7.56, "cost_score_avg": 4.65, "speed_score_avg": 7.0, "latency_score_avg": 1.07, "tool_calling_score_avg": 7.28, "judge_score_avg": 4.23, "tokens_per_second": 48.3, "latency_avg_s": 69.18, "total_input_tokens": 47383, "total_output_tokens": 463242 }, { "key": "claude-opus-4.7", "id": "anthropic/claude-opus-4-7", "name": "Claude Opus 4.7", "tier": "premium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 15.0, "cost_output_per_M": 75.0, "cost_per_1k_calls_usd": 117.0, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 254, "score_global": 6.54, "score_by_pillar": { "Contenido": 7.22, "Agentes": 6.97, "Coding": 7.31, "Razonamiento": 7.23 }, "score_by_suite": { "content_generation": 7.36, "tool_calling": 6.38, "task_management": 7.54, "code_generation": 7.48, "reasoning": 7.58, "summarization": 7.06, "presentation": 7.39, "startup_content": 7.42, "deep_reasoning": 6.85, "customer_support": 6.97, "structured_output": 7.49, "hallucination": 7.51, "creativity": 7.35, "string_precision": 7.43, "news_seo_writing": 6.5, "ocr_extraction": 6.88, "orchestration": 6.91, "multi_turn": 7.14, "policy_adherence": 7.02, "agent_capabilities": 7.01, "strategy": 7.38, "sales_outreach": 7.54, "translation": 7.35, "niah_es": 4.95 }, "quality_avg": 7.64, "cost_score_avg": 4.93, "speed_score_avg": 6.48, "latency_score_avg": 2.17, "tool_calling_score_avg": 7.19, "judge_score_avg": 4.22, "tokens_per_second": 51.9, "latency_avg_s": 12.68, "total_input_tokens": 9971743, "total_output_tokens": 189599 }, { "key": "qwen3.5-397b-cloud", "id": "qwen3.5:397b-cloud", "name": "Qwen 3.5 397B (Ollama Cloud)", "tier": "cloud_ollama", "provider": "ollama_cloud", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [ { "key": "ollama_cloud_pro", "name": "Ollama Cloud", "plan": "Pro", "price_month_usd": 30, "url": "https://ollama.com/cloud", "notes": "Rate limit varía por modelo. Recomendado para uso a volumen mid (1-10k calls/día)." } ], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 94, "score_global": 6.49, "score_by_pillar": { "Contenido": 6.04, "Agentes": 7.14, "Coding": 6.98, "Razonamiento": 5.57 }, "score_by_suite": { "translation": 4.41, "content_generation": 8.04, "tool_calling": 7.72, "task_management": 8.23, "code_generation": 8.28, "reasoning": 5.16, "summarization": 7.38, "presentation": 6.94, "startup_content": 7.04, "deep_reasoning": 4.89, "customer_support": 7.79, "structured_output": 6.69, "hallucination": 6.42, "creativity": 6.64, "string_precision": 7.71, "news_seo_writing": 4.39, "ocr_extraction": 5.31, "orchestration": 7.51, "multi_turn": 6.19, "policy_adherence": 6.44, "agent_capabilities": 6.44, "strategy": 6.46, "sales_outreach": 5.4 }, "quality_avg": 5.5, "cost_score_avg": 9.84, "speed_score_avg": 7.78, "latency_score_avg": 1.35, "tool_calling_score_avg": 7.24, "judge_score_avg": 4.05, "tokens_per_second": 75.4, "latency_avg_s": 32.0, "total_input_tokens": 35683, "total_output_tokens": 168538 }, { "key": "kimi-k2.6", "id": "moonshotai/kimi-k2.6", "name": "Kimi K2.6", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Modified MIT", "cost_input_per_M": 0.8, "cost_output_per_M": 3.5, "cost_per_1k_calls_usd": 5.49, "subscriptions": [], "notes": "Thinking model. Pesos públicos en HF (1.1T params)", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 182, "score_global": 6.4, "score_by_pillar": { "Contenido": 6.5, "Agentes": 6.23, "Razonamiento": 6.29, "Coding": 6.61 }, "score_by_suite": { "content_generation": 6.79, "tool_calling": 5.66, "task_management": 6.73, "reasoning": 6.91, "summarization": 5.81, "deep_reasoning": 5.88, "customer_support": 6.67, "structured_output": 6.76, "hallucination": 6.77, "string_precision": 6.91, "news_seo_writing": 5.72, "ocr_extraction": 5.96, "orchestration": 6.2, "multi_turn": 6.33, "policy_adherence": 6.32, "agent_capabilities": 5.93, "sales_outreach": 6.92, "translation": 6.56, "code_generation": 6.8, "presentation": 6.68, "startup_content": 6.85, "creativity": 6.65, "strategy": 6.03 }, "quality_avg": 7.68, "cost_score_avg": 4.85, "speed_score_avg": 5.13, "latency_score_avg": 1.1, "tool_calling_score_avg": 6.94, "judge_score_avg": 4.06, "tokens_per_second": 35.2, "latency_avg_s": 117.96, "total_input_tokens": 52060, "total_output_tokens": 556926 }, { "key": "qwen3.5-cloud", "id": "qwen3.5:cloud", "name": "Qwen 3.5 (Ollama Cloud default)", "tier": "cloud_ollama", "provider": "ollama_cloud", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [ { "key": "ollama_cloud_pro", "name": "Ollama Cloud", "plan": "Pro", "price_month_usd": 30, "url": "https://ollama.com/cloud", "notes": "Rate limit varía por modelo. Recomendado para uso a volumen mid (1-10k calls/día)." } ], "notes": "", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 91, "score_global": 6.4, "score_by_pillar": { "Contenido": 6.16, "Agentes": 6.82, "Coding": 6.11, "Razonamiento": 6.37 }, "score_by_suite": { "content_generation": 7.15, "tool_calling": 8.01, "task_management": 5.37, "code_generation": 6.7, "reasoning": 5.2, "summarization": 7.61, "presentation": 7.46, "startup_content": 6.13, "deep_reasoning": 6.41, "customer_support": 7.11, "structured_output": 6.33, "hallucination": 7.68, "creativity": 6.16, "string_precision": 6.82, "news_seo_writing": 4.23, "ocr_extraction": 4.61, "orchestration": 7.42, "multi_turn": 6.91, "policy_adherence": 5.82, "agent_capabilities": 6.64, "strategy": 6.15, "sales_outreach": 6.75, "translation": 5.69 }, "quality_avg": 5.29, "cost_score_avg": 10.0, "speed_score_avg": 7.52, "latency_score_avg": 1.32, "tool_calling_score_avg": 7.24, "judge_score_avg": 4.08, "tokens_per_second": 70.5, "latency_avg_s": 33.22, "total_input_tokens": 35995, "total_output_tokens": 154433 }, { "key": "nim-qwen3-next-thinking", "id": "qwen/qwen3-next-80b-a3b-thinking", "name": "Qwen 3-Next 80B Thinking (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Variante thinking — comparar con Qwen 3-Next Instruct", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 182, "score_global": 6.35, "score_by_pillar": { "Contenido": 6.43, "Agentes": 6.16, "Coding": 6.7, "Razonamiento": 6.12 }, "score_by_suite": { "content_generation": 6.76, "tool_calling": 6.46, "task_management": 6.22, "code_generation": 6.75, "reasoning": 6.11, "summarization": 6.44, "presentation": 6.49, "startup_content": 6.38, "deep_reasoning": 5.93, "customer_support": 5.88, "structured_output": 6.89, "hallucination": 6.59, "creativity": 6.47, "string_precision": 6.99, "news_seo_writing": 6.12, "ocr_extraction": 6.18, "orchestration": 5.52, "multi_turn": 6.6, "policy_adherence": 6.54, "agent_capabilities": 6.1, "strategy": 6.06, "sales_outreach": 6.71, "translation": 6.18 }, "quality_avg": 6.89, "cost_score_avg": 5.42, "speed_score_avg": 9.1, "latency_score_avg": 1.18, "tool_calling_score_avg": 6.47, "judge_score_avg": 3.69, "tokens_per_second": 102.5, "latency_avg_s": 17.43, "total_input_tokens": 61088, "total_output_tokens": 324760 }, { "key": "gemini-pro", "id": "google/gemini-2.5-pro", "name": "Gemini 2.5 Pro", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 1.25, "cost_output_per_M": 10.0, "cost_per_1k_calls_usd": 15.375, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 91, "score_global": 6.28, "score_by_pillar": { "Contenido": 6.34, "Agentes": 6.46, "Coding": 6.53, "Razonamiento": 5.52 }, "score_by_suite": { "content_generation": 7.36, "task_management": 6.65, "code_generation": 5.79, "reasoning": 5.74, "summarization": 6.24, "presentation": 6.22, "startup_content": 5.65, "deep_reasoning": 5.25, "customer_support": 7.0, "structured_output": 7.0, "hallucination": 6.22, "creativity": 6.77, "string_precision": 7.17, "news_seo_writing": 5.55, "ocr_extraction": 5.96, "orchestration": 6.58, "multi_turn": 6.99, "policy_adherence": 6.63, "agent_capabilities": 5.56, "strategy": 5.15, "sales_outreach": 6.71, "translation": 6.66, "tool_calling": 6.03 }, "quality_avg": 6.52, "cost_score_avg": 5.75, "speed_score_avg": 8.79, "latency_score_avg": 1.29, "tool_calling_score_avg": 7.07, "judge_score_avg": 3.76, "tokens_per_second": 91.2, "latency_avg_s": 18.53, "total_input_tokens": 24754, "total_output_tokens": 145620 }, { "key": "nim-step-3.5-flash", "id": "stepfun-ai/step-3.5-flash", "name": "Step 3.5 Flash (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Step 3.5 Flash - reemplazo de Step3 que fallo 91/91 con 404 en OpenRouter", "tested": true, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 103, "score_global": 6.2, "score_by_pillar": { "Contenido": 6.55, "Agentes": 6.33, "Coding": 6.01, "Razonamiento": 5.28 }, "score_by_suite": { "content_generation": 7.53, "tool_calling": 5.88, "task_management": 5.9, "code_generation": 6.33, "reasoning": 4.07, "summarization": 6.77, "presentation": 4.79, "startup_content": 7.16, "deep_reasoning": 4.84, "customer_support": 6.8, "structured_output": 6.4, "hallucination": 7.03, "creativity": 7.2, "string_precision": 7.04, "news_seo_writing": 6.1, "ocr_extraction": 4.19, "orchestration": 6.03, "multi_turn": 6.91, "policy_adherence": 6.93, "agent_capabilities": 5.96, "strategy": 5.64, "sales_outreach": 5.94, "translation": 5.73, "agent_long_horizon": 6.51 }, "quality_avg": 6.85, "cost_score_avg": 5.74, "speed_score_avg": 5.84, "latency_score_avg": 1.32, "tool_calling_score_avg": 6.92, "judge_score_avg": 4.17, "tokens_per_second": 41.3, "latency_avg_s": 69.33, "total_input_tokens": 249653, "total_output_tokens": 215660 }, { "key": "gemini-3.1-pro", "id": "google/gemini-3.1-pro-preview", "name": "Gemini 3.1 Pro", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 2.0, "cost_output_per_M": 12.0, "cost_per_1k_calls_usd": 18.6, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 175, "score_global": 6.18, "score_by_pillar": { "Contenido": 6.29, "Agentes": 6.5, "Coding": 6.35, "Razonamiento": 6.04 }, "score_by_suite": { "content_generation": 6.92, "tool_calling": 6.28, "task_management": 6.71, "code_generation": 6.26, "reasoning": 6.0, "summarization": 6.33, "presentation": 5.41, "startup_content": 6.61, "deep_reasoning": 5.75, "customer_support": 6.88, "structured_output": 6.58, "hallucination": 6.1, "creativity": 6.55, "string_precision": 6.83, "news_seo_writing": 5.3, "ocr_extraction": 5.64, "orchestration": 6.57, "multi_turn": 6.31, "policy_adherence": 6.4, "agent_capabilities": 6.4, "strategy": 6.59, "sales_outreach": 6.29, "translation": 6.81, "agent_long_horizon": 6.38, "niah_es": 5.95 }, "quality_avg": 7.4, "cost_score_avg": 2.95, "speed_score_avg": 8.26, "latency_score_avg": 1.55, "tool_calling_score_avg": 7.07, "judge_score_avg": 3.9, "tokens_per_second": 84.5, "latency_avg_s": 18.07, "total_input_tokens": 5984280, "total_output_tokens": 281979 }, { "key": "nim-kimi-k2-thinking", "id": "moonshotai/kimi-k2-thinking", "name": "Kimi K2 Thinking (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Modified MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Variante thinking de K2 — comparar con K2.6 thinking", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 102, "score_global": 6.17, "score_by_pillar": { "Contenido": 6.3, "Agentes": 6.27, "Coding": 6.1, "Razonamiento": 5.25 }, "score_by_suite": { "content_generation": 7.02, "tool_calling": 6.23, "task_management": 5.57, "code_generation": 5.25, "reasoning": 4.27, "summarization": 6.36, "presentation": 6.43, "startup_content": 6.2, "deep_reasoning": 4.84, "customer_support": 6.77, "structured_output": 6.88, "hallucination": 6.9, "creativity": 6.41, "string_precision": 6.99, "news_seo_writing": 5.84, "ocr_extraction": 5.1, "orchestration": 6.57, "multi_turn": 6.24, "policy_adherence": 6.87, "agent_capabilities": 5.58, "strategy": 5.43, "sales_outreach": 7.01, "translation": 5.34, "agent_long_horizon": 6.88 }, "quality_avg": 6.94, "cost_score_avg": 5.78, "speed_score_avg": 5.08, "latency_score_avg": 1.11, "tool_calling_score_avg": 6.86, "judge_score_avg": 4.07, "tokens_per_second": 30.5, "latency_avg_s": 78.4, "total_input_tokens": 200238, "total_output_tokens": 209595 }, { "key": "kimi-k2.5", "id": "moonshotai/kimi-k2.5", "name": "Kimi K2.5", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Modified MIT", "cost_input_per_M": 0.2, "cost_output_per_M": 0.8, "cost_per_1k_calls_usd": 1.26, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 130, "score_global": 6.14, "score_by_pillar": { "Contenido": 6.34, "Agentes": 6.31, "Coding": 6.18, "Razonamiento": 4.95 }, "score_by_suite": { "content_generation": 5.6, "tool_calling": 6.22, "task_management": 5.1, "code_generation": 5.96, "reasoning": 4.65, "summarization": 6.44, "presentation": 7.14, "startup_content": 6.57, "deep_reasoning": 3.96, "customer_support": 7.16, "structured_output": 7.35, "hallucination": 6.66, "creativity": 6.71, "string_precision": 7.09, "news_seo_writing": 5.64, "ocr_extraction": 4.51, "orchestration": 6.81, "multi_turn": 6.47, "policy_adherence": 6.54, "agent_capabilities": 6.42, "strategy": 5.82, "sales_outreach": 7.6, "translation": 5.74, "agent_long_horizon": 6.62 }, "quality_avg": 6.37, "cost_score_avg": 7.05, "speed_score_avg": 4.37, "latency_score_avg": 1.13, "tool_calling_score_avg": 7.12, "judge_score_avg": 3.98, "tokens_per_second": 29.0, "latency_avg_s": 93.31, "total_input_tokens": 228998, "total_output_tokens": 273771 }, { "key": "nim-kimi-k2.5", "id": "moonshotai/kimi-k2.5", "name": "Kimi K2.5 (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Modified MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Version nueva 2026, no tenemos K2.5 antes. Comparar con K2.6 thinking", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 130, "score_global": 6.14, "score_by_pillar": { "Contenido": 6.34, "Agentes": 6.31, "Coding": 6.18, "Razonamiento": 4.95 }, "score_by_suite": { "content_generation": 5.6, "tool_calling": 6.22, "task_management": 5.1, "code_generation": 5.96, "reasoning": 4.65, "summarization": 6.44, "presentation": 7.14, "startup_content": 6.57, "deep_reasoning": 3.96, "customer_support": 7.16, "structured_output": 7.35, "hallucination": 6.66, "creativity": 6.71, "string_precision": 7.09, "news_seo_writing": 5.64, "ocr_extraction": 4.51, "orchestration": 6.81, "multi_turn": 6.47, "policy_adherence": 6.54, "agent_capabilities": 6.42, "strategy": 5.82, "sales_outreach": 7.6, "translation": 5.74, "agent_long_horizon": 6.62 }, "quality_avg": 6.37, "cost_score_avg": 7.05, "speed_score_avg": 4.37, "latency_score_avg": 1.13, "tool_calling_score_avg": 7.12, "judge_score_avg": 3.98, "tokens_per_second": 29.0, "latency_avg_s": 93.31, "total_input_tokens": 228998, "total_output_tokens": 273771 }, { "key": "gpt-5.5", "id": "gpt-5.5", "name": "GPT-5.5", "tier": "premium", "provider": "openai_direct", "open_source": false, "license": "", "cost_input_per_M": 5.0, "cost_output_per_M": 30.0, "cost_per_1k_calls_usd": 46.5, "subscriptions": [], "notes": "", "tested": true, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 151, "score_global": 6.07, "score_by_pillar": { "Contenido": 6.67, "Agentes": 5.69, "Coding": 6.56, "Razonamiento": 6.51 }, "score_by_suite": { "translation": 6.79, "content_generation": 7.1, "tool_calling": 6.05, "task_management": 6.49, "code_generation": 6.64, "reasoning": 6.67, "summarization": 6.67, "presentation": 6.75, "startup_content": 6.64, "deep_reasoning": 6.19, "customer_support": 4.25, "structured_output": 6.93, "hallucination": 6.92, "creativity": 6.76, "string_precision": 6.53, "news_seo_writing": 5.84, "ocr_extraction": 6.26, "orchestration": 5.37, "multi_turn": 6.37, "policy_adherence": 6.59, "agent_capabilities": 5.15, "sales_outreach": 7.15, "strategy": 6.56, "agent_long_horizon": 6.51, "niah_es_lite": 5.44 }, "quality_avg": 7.29, "cost_score_avg": 2.75, "speed_score_avg": 6.74, "latency_score_avg": 2.78, "tool_calling_score_avg": 7.16, "judge_score_avg": 4.13, "tokens_per_second": 53.0, "latency_avg_s": 24.12, "total_input_tokens": 1391651, "total_output_tokens": 203823 }, { "key": "deepseek-v4-pro", "id": "deepseek/deepseek-v4-pro", "name": "DeepSeek V4 Pro", "tier": "medium", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 1.74, "cost_output_per_M": 3.48, "cost_per_1k_calls_usd": 5.742, "subscriptions": [], "notes": "1.6T params, 49B activos, 1M context. Flagship V4.", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 126, "score_global": 5.62, "score_by_pillar": { "Contenido": 6.14, "Agentes": 6.42, "Coding": 6.42, "Razonamiento": 6.22 }, "score_by_suite": { "content_generation": 6.3, "tool_calling": 6.63, "task_management": 6.94, "code_generation": 7.11, "reasoning": 5.61, "summarization": 6.66, "presentation": 4.62, "startup_content": 5.77, "deep_reasoning": 5.88, "structured_output": 6.84, "hallucination": 6.88, "creativity": 7.04, "string_precision": 6.84, "news_seo_writing": 5.94, "ocr_extraction": 5.25, "orchestration": 6.83, "multi_turn": 6.43, "policy_adherence": 5.87, "agent_capabilities": 6.18, "strategy": 6.85, "sales_outreach": 7.1, "translation": 5.53, "agent_long_horizon": 6.81, "niah_es_lite": 4.29 }, "quality_avg": 6.32, "cost_score_avg": 4.73, "speed_score_avg": 4.55, "latency_score_avg": 1.22, "tool_calling_score_avg": 7.04, "judge_score_avg": 3.03, "tokens_per_second": 26.8, "latency_avg_s": 69.99, "total_input_tokens": 1540300, "total_output_tokens": 240893 }, { "key": "deepseek-v4-flash-cloud", "id": "deepseek-v4-flash", "name": "DeepSeek V4 Flash (Ollama Cloud)", "tier": "cloud_ollama", "provider": "ollama_cloud", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [ { "key": "ollama_cloud_pro", "name": "Ollama Cloud", "plan": "Pro", "price_month_usd": 30, "url": "https://ollama.com/cloud", "notes": "Rate limit varía por modelo. Recomendado para uso a volumen mid (1-10k calls/día)." } ], "notes": "Variante mas chica de V4. Comparar con V4 Flash NIM (7.07 score).", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 57, "score_global": 4.96, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.79, "niah_es_lite": 4.47 }, "quality_avg": 5.15, "cost_score_avg": 4.26, "speed_score_avg": 4.44, "latency_score_avg": 2.0, "tool_calling_score_avg": 7.0, "judge_score_avg": 1.46, "tokens_per_second": 29.2, "latency_avg_s": 48.31, "total_input_tokens": 1478714, "total_output_tokens": 68911 }, { "key": "deepseek-v4-pro-cloud", "id": "deepseek-v4-pro", "name": "DeepSeek V4 Pro (Ollama Cloud)", "tier": "cloud_ollama", "provider": "ollama_cloud", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [ { "key": "ollama_cloud_pro", "name": "Ollama Cloud", "plan": "Pro", "price_month_usd": 30, "url": "https://ollama.com/cloud", "notes": "Rate limit varía por modelo. Recomendado para uso a volumen mid (1-10k calls/día)." } ], "notes": "Recien agregado a Ollama Cloud (abril 28). Smoke test 22s OK, sin timeouts vs OpenRouter (76% cobertura) y NIM (504s timeouts).", "tested": true, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 55, "score_global": 4.86, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 7.27, "niah_es_lite": 4.32 }, "quality_avg": 5.24, "cost_score_avg": 4.23, "speed_score_avg": 3.16, "latency_score_avg": 1.35, "tool_calling_score_avg": 7.0, "judge_score_avg": 1.44, "tokens_per_second": 18.6, "latency_avg_s": 91.49, "total_input_tokens": 1399556, "total_output_tokens": 65064 }, { "key": "deepseek-v4-flash", "id": "deepseek/deepseek-v4-flash", "name": "DeepSeek V4 Flash (OpenRouter)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 0.14, "cost_output_per_M": 0.28, "cost_per_1k_calls_usd": 0.462, "subscriptions": [], "notes": "284B params, 13B activos, 1M context. Sucesor V3.2.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 3, "score_global": 8.26, "score_by_pillar": { "Contenido": 8.26 }, "score_by_suite": { "content_generation": 8.26 }, "quality_avg": 9.16, "cost_score_avg": 9.0, "speed_score_avg": 8.0, "latency_score_avg": 1.33, "tool_calling_score_avg": 7.0, "judge_score_avg": 4.73, "tokens_per_second": 66.7, "latency_avg_s": 13.41, "total_input_tokens": 220, "total_output_tokens": 2682 }, { "key": "gemini-3.1-flash-lite-thinking", "id": "google/gemini-3.1-flash-lite-preview", "name": "Gemini 3.1 Flash Lite (thinking)", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.25, "cost_output_per_M": 1.5, "cost_per_1k_calls_usd": 2.325, "subscriptions": [], "notes": "Misma versión con thinking forzado.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 12, "score_global": 7.46, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 7.46 }, "quality_avg": 9.0, "cost_score_avg": 4.0, "speed_score_avg": 9.5, "latency_score_avg": 3.0, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 203.0, "latency_avg_s": 41.13, "total_input_tokens": 146170, "total_output_tokens": 102530 }, { "key": "gemini-flash-thinking", "id": "google/gemini-2.5-flash", "name": "Gemini 2.5 Flash (thinking)", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0.3, "cost_output_per_M": 2.5, "cost_per_1k_calls_usd": 3.84, "subscriptions": [], "notes": "Misma versión que gemini-flash con thinking forzado (effort=high).", "tested": false, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 12, "score_global": 7.27, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 7.27 }, "quality_avg": 8.92, "cost_score_avg": 3.83, "speed_score_avg": 8.83, "latency_score_avg": 3.25, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 136.0, "latency_avg_s": 86.41, "total_input_tokens": 186429, "total_output_tokens": 111130 }, { "key": "nim-deepseek-v4-pro", "id": "deepseek-ai/deepseek-v4-pro", "name": "DeepSeek V4 Pro (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Re-test del flagship via NIM (vs OpenRouter que dio 76% cobertura en Lote 7-8)", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 3, "score_global": 7.26, "score_by_pillar": { "Contenido": 7.37, "Agentes": 7.04 }, "score_by_suite": { "content_generation": 7.37, "tool_calling": 7.04 }, "quality_avg": 8.83, "cost_score_avg": 7.0, "speed_score_avg": 1.33, "latency_score_avg": 1.0, "tool_calling_score_avg": 8.0, "judge_score_avg": 4.67, "tokens_per_second": 4.5, "latency_avg_s": 176.92, "total_input_tokens": 881, "total_output_tokens": 1584 }, { "key": "minimax-m2.7-direct", "id": "MiniMax-M2.7", "name": "MiniMax M2.7 (directo)", "tier": "cheap", "provider": "minimax_direct", "open_source": false, "license": "", "cost_input_per_M": 0.3, "cost_output_per_M": 1.2, "cost_per_1k_calls_usd": 1.89, "subscriptions": [ { "key": "minimax_agent_pro", "name": "MiniMax Agent Pro", "plan": "Agent Pro", "price_month_usd": 19, "url": "https://api.minimax.io", "notes": "Acceso a M2.7 highspeed + límites generosos para agentes (1k+ calls/día)." } ], "notes": "API directa MiniMax. Disponible también en sub Agent Pro $19/mes.", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 39, "score_global": 7.19, "score_by_pillar": { "Contenido": 7.37, "Agentes": 6.85, "Coding": 7.91, "Razonamiento": 7.68 }, "score_by_suite": { "content_generation": 7.32, "tool_calling": 6.32, "task_management": 7.56, "code_generation": 7.91, "reasoning": 7.68, "summarization": 6.71, "presentation": 7.61, "startup_content": 7.58, "agent_long_horizon": 6.84 }, "quality_avg": 8.5, "cost_score_avg": 5.87, "speed_score_avg": 6.72, "latency_score_avg": 1.64, "tool_calling_score_avg": 7.18, "judge_score_avg": null, "tokens_per_second": 52.0, "latency_avg_s": 52.21, "total_input_tokens": 142391, "total_output_tokens": 86778 }, { "key": "kimi-k2.5-thinking", "id": "moonshotai/kimi-k2.5", "name": "Kimi K2.5 (thinking)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Modified MIT", "cost_input_per_M": 0.2, "cost_output_per_M": 0.8, "cost_per_1k_calls_usd": 1.26, "subscriptions": [], "notes": "Misma versión que kimi-k2.5 pero con reasoning forzado vía OpenRouter (effort=high).", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 12, "score_global": 7.16, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 7.16 }, "quality_avg": 9.12, "cost_score_avg": 4.67, "speed_score_avg": 5.17, "latency_score_avg": 1.08, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 36.4, "latency_avg_s": 318.08, "total_input_tokens": 217776, "total_output_tokens": 113040 }, { "key": "minimax-m2.7-highspeed", "id": "MiniMax-M2.7-highspeed", "name": "MiniMax M2.7 Highspeed", "tier": "cheap", "provider": "minimax_direct", "open_source": false, "license": "", "cost_input_per_M": 0.3, "cost_output_per_M": 1.2, "cost_per_1k_calls_usd": 1.89, "subscriptions": [ { "key": "minimax_agent_pro", "name": "MiniMax Agent Pro", "plan": "Agent Pro", "price_month_usd": 19, "url": "https://api.minimax.io", "notes": "Acceso a M2.7 highspeed + límites generosos para agentes (1k+ calls/día)." } ], "notes": "Acceso vía sub Agent Pro $19/mes. Misma calidad que M2.7 directo, latencia ultra baja.", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 39, "score_global": 7.14, "score_by_pillar": { "Contenido": 7.46, "Agentes": 6.87, "Coding": 7.64, "Razonamiento": 7.85 }, "score_by_suite": { "content_generation": 7.34, "tool_calling": 6.22, "task_management": 7.73, "code_generation": 7.64, "reasoning": 7.85, "summarization": 7.4, "presentation": 7.63, "startup_content": 7.51, "agent_long_horizon": 6.61 }, "quality_avg": 8.36, "cost_score_avg": 5.97, "speed_score_avg": 6.97, "latency_score_avg": 1.69, "tool_calling_score_avg": 7.18, "judge_score_avg": null, "tokens_per_second": 49.3, "latency_avg_s": 54.47, "total_input_tokens": 171687, "total_output_tokens": 91414 }, { "key": "nim-mistral-nemotron", "id": "mistralai/mistral-nemotron", "name": "Mistral-Nemotron (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Colaboración Mistral × NVIDIA, optimizado en Nemo", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 12, "score_global": 7.05, "score_by_pillar": { "Contenido": 7.72, "Agentes": 6.53, "Coding": 7.93 }, "score_by_suite": { "content_generation": 7.72, "tool_calling": 5.69, "task_management": 7.66, "code_generation": 7.93 }, "quality_avg": 8.04, "cost_score_avg": 8.0, "speed_score_avg": 5.83, "latency_score_avg": 3.17, "tool_calling_score_avg": 5.5, "judge_score_avg": 4.1, "tokens_per_second": 38.3, "latency_avg_s": 19.95, "total_input_tokens": 3013, "total_output_tokens": 5806 }, { "key": "claude-haiku-4.5", "id": "anthropic/claude-haiku-4.5", "name": "Claude Haiku 4.5", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 1.0, "cost_output_per_M": 5.0, "cost_per_1k_calls_usd": 7.8, "subscriptions": [], "notes": "Anthropic Haiku family. Hybrid (extended thinking opt-in).", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 12, "score_global": 6.98, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.98 }, "quality_avg": 8.38, "cost_score_avg": 3.67, "speed_score_avg": 8.83, "latency_score_avg": 2.83, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 102.4, "latency_avg_s": 44.33, "total_input_tokens": 196551, "total_output_tokens": 52340 }, { "key": "hermes-4-70b-thinking", "id": "nousresearch/hermes-4-70b", "name": "Hermes 4 70B (thinking)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Llama 3 community", "cost_input_per_M": 0.13, "cost_output_per_M": 0.4, "cost_per_1k_calls_usd": 0.639, "subscriptions": [], "notes": "Misma versión que hermes-4-70b pero con reasoning forzado vía OpenRouter (effort=high). Technical report reporta +12 puntos MMLU al activar reasoning (76.7 → 88.4).", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 12, "score_global": 6.84, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.84 }, "quality_avg": 8.79, "cost_score_avg": 2.33, "speed_score_avg": 8.0, "latency_score_avg": 1.42, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 73.1, "latency_avg_s": 95.39, "total_input_tokens": 301519, "total_output_tokens": 86335 }, { "key": "claude-haiku-4.5-thinking", "id": "anthropic/claude-haiku-4.5", "name": "Claude Haiku 4.5 (thinking)", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 1.0, "cost_output_per_M": 5.0, "cost_per_1k_calls_usd": 7.8, "subscriptions": [], "notes": "Misma versión con extended thinking forzado.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 12, "score_global": 6.58, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.58 }, "quality_avg": 7.79, "cost_score_avg": 3.17, "speed_score_avg": 9.17, "latency_score_avg": 2.25, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 116.9, "latency_avg_s": 141.26, "total_input_tokens": 255657, "total_output_tokens": 107160 }, { "key": "claude-sonnet-4.6-thinking", "id": "anthropic/claude-sonnet-4-6", "name": "Claude Sonnet 4.6 (thinking)", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 3.0, "cost_output_per_M": 15.0, "cost_per_1k_calls_usd": 23.4, "subscriptions": [], "notes": "Mismo modelo que claude-sonnet-4.6 con extended thinking forzado (effort=high). Reasoning tokens facturados como output a $15/M — ~5-7x más caro por test que sin thinking.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 12, "score_global": 6.49, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.49 }, "quality_avg": 8.75, "cost_score_avg": 1.17, "speed_score_avg": 7.33, "latency_score_avg": 1.33, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 57.7, "latency_avg_s": 149.09, "total_input_tokens": 306217, "total_output_tokens": 97264 }, { "key": "gemini-3.1-pro-thinking", "id": "google/gemini-3.1-pro-preview", "name": "Gemini 3.1 Pro (thinking)", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 2.0, "cost_output_per_M": 12.0, "cost_per_1k_calls_usd": 18.6, "subscriptions": [], "notes": "Misma versión con thinking forzado (effort=high). Costo ~5x mayor por test que sin thinking.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 12, "score_global": 6.42, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.42 }, "quality_avg": 8.5, "cost_score_avg": 1.67, "speed_score_avg": 8.17, "latency_score_avg": 1.33, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 83.8, "latency_avg_s": 133.43, "total_input_tokens": 167761, "total_output_tokens": 111329 }, { "key": "kimi-k2.6-thinking", "id": "moonshotai/kimi-k2.6", "name": "Kimi K2.6 (thinking)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Modified MIT", "cost_input_per_M": 0.8, "cost_output_per_M": 3.5, "cost_per_1k_calls_usd": 5.49, "subscriptions": [], "notes": "Misma versión que kimi-k2.6 con thinking forzado vía OpenRouter (effort=high).", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 12, "score_global": 6.25, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.25 }, "quality_avg": 8.5, "cost_score_avg": 1.33, "speed_score_avg": 5.67, "latency_score_avg": 1.17, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 40.0, "latency_avg_s": 494.23, "total_input_tokens": 197721, "total_output_tokens": 128021 }, { "key": "hermes-4-405b-thinking", "id": "nousresearch/hermes-4-405b", "name": "Hermes 4 405B (thinking)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Llama 3 community", "cost_input_per_M": 1.0, "cost_output_per_M": 3.0, "cost_per_1k_calls_usd": 4.8, "subscriptions": [], "notes": "Misma versión que hermes-4-405b pero con reasoning forzado.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 12, "score_global": 6.24, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.24 }, "quality_avg": 7.92, "cost_score_avg": 2.5, "speed_score_avg": 6.0, "latency_score_avg": 1.08, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 35.6, "latency_avg_s": 188.98, "total_input_tokens": 265460, "total_output_tokens": 76894 }, { "key": "claude-opus-4.7-thinking", "id": "anthropic/claude-opus-4-7", "name": "Claude Opus 4.7 (thinking)", "tier": "premium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 15.0, "cost_output_per_M": 75.0, "cost_per_1k_calls_usd": 117.0, "subscriptions": [], "notes": "Mismo modelo que claude-opus-4.7 con extended thinking forzado (effort=high). Reasoning tokens facturados como output a $75/M — ~5-7x más caro por test que sin thinking.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 12, "score_global": 6.01, "score_by_pillar": {}, "score_by_suite": { "agent_long_horizon": 6.01 }, "quality_avg": 8.38, "cost_score_avg": 1.0, "speed_score_avg": 7.83, "latency_score_avg": 1.25, "tool_calling_score_avg": 7.0, "judge_score_avg": null, "tokens_per_second": 63.3, "latency_avg_s": 204.26, "total_input_tokens": 374227, "total_output_tokens": 113625 }, { "key": "nim-magistral-small", "id": "mistralai/magistral-small-2506", "name": "Mistral Magistral Small (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Mistral con razonamiento, tamano medio", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 10, "score_global": 5.51, "score_by_pillar": { "Contenido": 7.21, "Agentes": 4.38 }, "score_by_suite": { "content_generation": 7.21, "tool_calling": 5.31, "task_management": 2.54 }, "quality_avg": 5.9, "cost_score_avg": 6.6, "speed_score_avg": 4.0, "latency_score_avg": 2.1, "tool_calling_score_avg": 5.2, "judge_score_avg": 4.56, "tokens_per_second": 23.8, "latency_avg_s": 43.64, "total_input_tokens": 4528, "total_output_tokens": 10999 }, { "key": "deepseek-r1-free", "id": "deepseek/deepseek-r1:free", "name": "DeepSeek R1 (free)", "tier": "free", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "llama-3.3-70b-free", "id": "meta-llama/llama-3.3-70b-instruct:free", "name": "Llama 3.3 70B (free)", "tier": "free", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "qwen3-coder-free", "id": "qwen/qwen3-coder-480b:free", "name": "Qwen3 Coder 480B (free)", "tier": "free", "provider": "openrouter", "open_source": true, "license": "", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "deepseek-v3", "id": "deepseek-v3", "name": "DeepSeek V3 (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "qwen-3.5-plus", "id": "qwen/qwen3.5-plus", "name": "Qwen 3.5 Plus", "tier": "cheap", "provider": "openrouter", "open_source": false, "license": "Proprietary", "cost_input_per_M": 1.2, "cost_output_per_M": 2.0, "cost_per_1k_calls_usd": 3.36, "subscriptions": [], "notes": "Plus = API-only propietario de Alibaba", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "gpt-4o", "id": "openai/gpt-4o", "name": "GPT-4o", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 2.5, "cost_output_per_M": 10.0, "cost_per_1k_calls_usd": 15.75, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "claude-sonnet", "id": "anthropic/claude-sonnet-4", "name": "Claude Sonnet 4", "tier": "medium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 3.0, "cost_output_per_M": 15.0, "cost_per_1k_calls_usd": 23.4, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "gpt-5.5-pro", "id": "gpt-5.5-pro", "name": "GPT-5.5 Pro", "tier": "premium", "provider": "openai_responses", "open_source": false, "license": "", "cost_input_per_M": 30.0, "cost_output_per_M": 180.0, "cost_per_1k_calls_usd": 279.0, "subscriptions": [], "notes": "Sólo /v1/responses (no chat/completions)", "tested": false, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "groq-gpt-oss-120b", "id": "openai/gpt-oss-120b", "name": "GPT-OSS 120B (Groq)", "tier": "cheap", "provider": "groq_direct", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.15, "cost_output_per_M": 0.6, "cost_per_1k_calls_usd": 0.945, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "gpt-4o-high", "id": "openai/gpt-4o:high", "name": "GPT-4o High", "tier": "premium", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 5.0, "cost_output_per_M": 15.0, "cost_per_1k_calls_usd": 24.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": true, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "nim-nemotron-ultra-253b", "id": "nvidia/llama-3.1-nemotron-ultra-253b-v1", "name": "Nemotron Ultra 253B (NIM)", "tier": "cloud_nim", "provider": "nvidia_nim", "open_source": true, "license": "NVIDIA Open Model", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "Modelo más grande de la familia Nemotron", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "mimo-v2-flash-free", "id": "xiaomi/mimo-v2-flash:free", "name": "MiMo-V2-Flash (free)", "tier": "free", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "qwen-3.6-plus-free", "id": "qwen/qwen3.6-plus:free", "name": "Qwen 3.6 Plus (free)", "tier": "free", "provider": "openrouter", "open_source": false, "license": "", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "step3", "id": "stepfun-ai/step3", "name": "Step3 (StepFun)", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 1.0, "cost_output_per_M": 3.0, "cost_per_1k_calls_usd": 4.8, "subscriptions": [], "notes": "MoE 321B/38B activos, multimodal reasoning. 65K context. Lanzado ago 2025.", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "seed-oss-36b", "id": "bytedance/seed-oss-36b-instruct", "name": "Seed-OSS 36B Instruct", "tier": "cheap", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0.2, "cost_output_per_M": 0.6, "cost_per_1k_calls_usd": 0.96, "subscriptions": [], "notes": "Reasoning + math + coding + agentic. 131K context. ByteDance.", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "local-nemotron-3-nano-omni-reasoning", "id": "hf.co/unsloth/NVIDIA-Nemotron-3-Nano-Omni-30B-A3B-Reasoning-GGUF:Q4_K_M", "name": "Nemotron 3 Nano Omni 30B-A3B Reasoning (DGX Spark Q4_K_M)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "NVIDIA Open License", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "MoE 30B totales / 3B activos (A3B), thinking + multimodal. GGUF community por unsloth — los tags oficiales de Ollama (nemotron3:33b) son la versión base, no la Omni Reasoning. Ollama: `ollama pull hf.co/unsloth/NVIDIA-Nemotron-3-Nano-Omni-30B-A3B-Reasoning-GGUF:Q4_K_M`. Comparar contra el mismo modelo via NIM (FP16) para medir costo de cuantización.", "tested": false, "tool_calling": true, "thinking": true, "multimodal": true, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "qwen3.5-25b", "id": "qwen3.5:25b", "name": "Qwen 3.5 25B (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "qwen3.5-72b", "id": "qwen3.5:72b", "name": "Qwen 3.5 72B (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "llama3.3-70b", "id": "llama3.3:70b", "name": "Llama 3.3 70B (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Llama Community", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "llama4-maverick", "id": "llama4-maverick", "name": "Llama 4 Maverick (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Llama Community", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "gemma4-26b-moe", "id": "gemma4:26b", "name": "Gemma 4 26B MoE (local, solo 3.8B activos)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": true, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "mistral-nemo-12b", "id": "mistral-nemo", "name": "Mistral Nemo 12B (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "Apache 2.0", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "phi-4-14b", "id": "phi4", "name": "Phi-4 14B (local)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": false, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 }, { "key": "minimax-m2.5", "id": "MiniMax-M2.5", "name": "MiniMax M2.5 (local, open-source)", "tier": "local", "provider": "openrouter", "open_source": true, "license": "MIT", "cost_input_per_M": 0, "cost_output_per_M": 0, "cost_per_1k_calls_usd": 0.0, "subscriptions": [], "notes": "", "tested": false, "tool_calling": true, "thinking": false, "multimodal": false, "runs": 0, "score_global": null, "score_by_pillar": {}, "score_by_suite": {}, "judge_score_avg": null, "tokens_per_second": null, "latency_avg_s": null, "total_input_tokens": 0, "total_output_tokens": 0 } ] }