Benchmarks
Clasificación
Todos los modelos de texto activos a los que podemos llegar, con lo que tardan en responder y con qué frecuencia ganan un duelo. No se oculta nada: un modelo sin medir conserva su fila, y esa fila explica por qué.
- Caso normal
- Lo que tarda una respuesta corriente. La mitad de las llamadas volvió antes y la otra mitad después — así que es lo que cabe esperar en una petición normal.
- Caso lento
- Lo malo que se pone en un mal día. Aproximadamente una llamada de cada veinte es más lenta que esto. Un modelo con caso normal bajo pero caso lento alto es rápido hasta que deja de serlo.
- Tasa de victoria
- De 100 duelos frente a un modelo medio de esta tabla, cuántos gana este. 50 es la media, más alto es mejor. Se calcula con todas las comparaciones evaluadas, no con una puntuación suelta.
105 de 110 modelos medidos por completo · 0 cronometrados una vez · 0 a la espera de una primera pasada · 5 que no podemos medir.
Filtro:
| # | |||||||
|---|---|---|---|---|---|---|---|
| 1 | Mistral-Nemo-Instruct-2407C | OVH AI Endpoints (GRA) | C | 109 ms | 204 ms | 30% | 2026-10-30 |
| 2 | Mistral-Small-3.2-24B-Instruct-2506B | OVH AI Endpoints (GRA) | B | 121 ms | 2.2 s | 50% | 2026-10-30 |
| 3 | Qwen2.5-VL-72B-InstructB | OVH AI Endpoints (GRA) | B | 165 ms | 337 ms | 53% | 2026-10-30 |
| 4 | Nous Hermes 3 70BA | OpenRouter | A | 172 ms | 188 ms | sin comparar | 2026-10-30 |
| 5 | NVIDIA Nemotron Super 49B v1.5A | OpenRouter | A | 173 ms | 199 ms | sin comparar | 2026-07-17 |
| 6 | DeepSeek v3.2A | OpenRouter | A | 185 ms | 306 ms | sin comparar | 2026-10-30 |
| 7 | Llama 4 ScoutA | OpenRouter | A | 308 ms | 336 ms | sin comparar | 2026-10-30 |
| 8 | Qwen3.5-397B-A17BA | OVH AI Endpoints (GRA) | A | 326 ms | 376 ms | 29% | 2026-10-30 |
| 9 | Meta-Llama-3_3-70B-InstructB | OVH AI Endpoints (GRA) | B | 353 ms | 5.2 s | 53% | 2026-10-30 |
| 10 | Cohere Command-AA | OpenRouter | A | 356 ms | 4.7 s | sin comparar | 2026-10-30 |
| 11 | Qwen 2.5 VL 72B InstructA | OpenRouter | A | 440 ms | 1.2 s | sin comparar | 2026-10-30 |
| 12 | Gemini 2.5 Flash-LiteB | Google Gemini | B | 457 ms | 474 ms | 49% | 2026-10-30 |
| 13 | Gemini 3.1 Flash LiteB | Google Gemini | B | 458 ms | 677 ms | 60% | 2026-10-30 |
| 14 | DeepSeek v4 ProA | OpenRouter | A | 459 ms | 553 ms | sin comparar | 2026-10-30 |
| 15 | Gemini 3.5 Flash-LiteB | Google Gemini | B | 462 ms | 524 ms | 72% | 2026-10-30 |
| 16 | gpt-5.4-nano-2026-03-17A | OpenAI | A | 475 ms | 664 ms | 57% | 2026-10-30 |
| 17 | Llama 3.3 70B InstructA | OpenRouter | A | 486 ms | 499 ms | sin comparar | 2026-10-30 |
| 18 | Gemini Flash-Lite LatestC | Google Gemini | C | 491 ms | 516 ms | 63% | 2026-10-30 |
| 19 | gpt-4o-mini-2024-07-18C | OpenAI | C | 505 ms | 535 ms | 55% | 2026-10-30 |
| 20 | MiniMax M2.5A | OpenRouter | A | 525 ms | 794 ms | sin comparar | 2026-10-30 |
| 21 | Claude Haiku 4.5A | Anthropic | A | 527 ms | 543 ms | 52% | 2026-10-30 |
| 22 | gpt-4o-miniC | OpenAI | C | 527 ms | 717 ms | 57% | 2026-10-30 |
| 23 | Gemini 2.5 FlashA | Google Gemini | A | 529 ms | 1.5 s | 16% | 2026-10-30 |
| 24 | gpt-5.4-nanoC | OpenAI | C | 536 ms | 777 ms | 52% | 2026-10-30 |
| 25 | Grok 4.20 Non-Reasoning (0309)B | xAI | B | 553 ms | 597 ms | sin comparar | 2026-10-30 |
| 26 | gpt-4oC | OpenAI | C | 554 ms | 844 ms | 62% | 2026-10-30 |
| 27 | gpt-4o-2024-08-06C | OpenAI | C | 565 ms | 590 ms | 61% | 2026-10-30 |
| 28 | gpt-oss-120bC | OVH AI Endpoints (GRA) | C | 566 ms | 799 ms | 60% | 2026-10-30 |
| 29 | gpt-4.1B | OpenAI | B | 574 ms | 678 ms | 66% | 2026-10-30 |
| 30 | gpt-4.1-2025-04-14C | OpenAI | C | 589 ms | 728 ms | 70% | 2026-10-30 |
| 31 | Llama 4 MaverickA | OpenRouter | A | 595 ms | 602 ms | sin comparar | 2026-10-30 |
| 32 | gpt-5.2B | OpenAI | B | 598 ms | 1.0 s | 69% | 2026-10-30 |
| 33 | gpt-4.1-mini-2025-04-14C | OpenAI | C | 599 ms | 628 ms | 68% | 2026-10-30 |
| 34 | o3-2025-04-16B | OpenAI | B | 607 ms | 898 ms | 61% | 2026-10-30 |
| 35 | Gemini 3 Flash PreviewC | Google Gemini | C | 621 ms | 704 ms | 44% | 2026-10-30 |
| 36 | gpt-5-mini-2025-08-07B | OpenAI | B | 637 ms | 1.1 s | 53% | 2026-10-30 |
| 37 | gpt-4.1-miniC | OpenAI | C | 644 ms | 670 ms | 72% | 2026-10-30 |
| 38 | gpt-5.4-miniA | OpenAI | A | 666 ms | 675 ms | 65% | 2026-10-30 |
| 39 | gpt-5.2-2025-12-11B | OpenAI | B | 675 ms | 1.0 s | 69% | 2026-10-30 |
| 40 | Claude Opus 4.7B | Anthropic | B | 680 ms | 718 ms | 64% | 2026-10-30 |
| 41 | Claude Opus 4.5B | Anthropic | B | 698 ms | 729 ms | 63% | 2026-10-30 |
| 42 | Qwen 3.6 PlusA | OpenRouter | A | 700 ms | 784 ms | sin comparar | 2026-10-30 |
| 43 | o3C | OpenAI | C | 740 ms | 822 ms | 64% | 2026-10-30 |
| 44 | Claude Sonnet 4.5B | Anthropic | B | 747 ms | 762 ms | 60% | 2026-10-30 |
| 45 | gpt-5.1B | OpenAI | B | 750 ms | 758 ms | 71% | 2026-10-30 |
| 46 | gpt-5.6-lunaB | OpenAI | B | 752 ms | 951 ms | 65% | 2026-10-30 |
| 47 | Claude Sonnet 4.6A | Anthropic | A | 768 ms | 918 ms | 59% | 2026-10-30 |
| 48 | gpt-5-2025-08-07B | OpenAI | B | 775 ms | 819 ms | 42% | 2026-10-30 |
| 49 | gpt-5C | OpenAI | C | 780 ms | 885 ms | 44% | 2026-10-30 |
| 50 | gpt-5-nano-2025-08-07B | OpenAI | B | 817 ms | 1.0 s | 32% | 2026-10-30 |
| 51 | gpt-5.4A | OpenAI | A | 818 ms | 1.0 s | 66% | 2026-10-30 |
| 52 | gpt-5.6-solB | OpenAI | B | 838 ms | 1.2 s | 59% | 2026-10-30 |
| 53 | gpt-5-nanoC | OpenAI | C | 838 ms | 886 ms | 26% | 2026-10-30 |
| 54 | Grok 4.7A | xAI | A | 845 ms | 1.1 s | sin comparar | 2026-10-30 |
| 55 | gpt-6-solA | OpenAI | A | 853 ms | 988 ms | 56% | 2026-10-30 |
| 56 | gpt-5-miniC | OpenAI | C | 855 ms | 875 ms | 56% | 2026-10-30 |
| 57 | Gemini 3.6 FlashB | Google Gemini | B | 869 ms | 874 ms | 15% | 2026-10-30 |
| 58 | gpt-6-lunaB | OpenAI | B | 883 ms | 924 ms | 65% | 2026-10-30 |
| 59 | gpt-5.1-2025-11-13B | OpenAI | B | 908 ms | 911 ms | 68% | 2026-10-30 |
| 60 | Claude Opus 5A | Anthropic | A | 923 ms | 933 ms | 52% | 2026-10-30 |
| 61 | Gemini 3.5 FlashA | Google Gemini | A | 939 ms | 968 ms | 27% | 2026-10-30 |
| 62 | gpt-4o-2024-11-20C | OpenAI | C | 961 ms | 1.1 s | 60% | 2026-10-30 |
| 63 | Grok 4.5B | xAI | B | 976 ms | 1.1 s | sin comparar | 2026-10-30 |
| 64 | Claude Opus 4.8A | Anthropic | A | 983 ms | 1.0 s | 60% | 2026-10-30 |
| 65 | Claude Sonnet 5A | Anthropic | A | 989 ms | 1.1 s | 58% | 2026-10-30 |
| 66 | gpt-5.6-terraB | OpenAI | B | 1.0 s | 1.2 s | 72% | 2026-10-30 |
| 67 | gpt-5.4-mini-2026-03-17A | OpenAI | A | 1.1 s | 1.6 s | 65% | 2026-10-30 |
| 68 | gpt-6.1-solA | OpenAI | A | 1.1 s | 1.2 s | 68% | 2026-10-30 |
| 69 | Grok 4.20 Reasoning (0309)B | xAI | B | 1.1 s | 1.4 s | sin comparar | 2026-10-30 |
| 70 | gpt-5.5C | OpenAI | C | 1.1 s | 1.1 s | 71% | 2026-10-30 |
| 71 | Qwen 3.7 MaxA | OpenRouter | A | 1.1 s | 1.2 s | sin comparar | 2026-10-30 |
| 72 | gpt-5.4-2026-03-05B | OpenAI | B | 1.1 s | 1.1 s | 68% | 2026-10-30 |
| 73 | gpt-6-astraA | OpenAI | A | 1.1 s | 1.6 s | 70% | 2026-10-30 |
| 74 | GLM-4.5V (vision)A | Z.ai (GLM / Zhipu) | A | 1.1 s | 1.5 s | 29% | 2026-10-30 |
| 75 | Claude Fable 5.1A | Anthropic | A | 1.2 s | 1.2 s | sin comparar | 2026-10-30 |
| 76 | Gemini 2.5 ProA | Google Gemini | A | 1.2 s | 2.3 s | 18% | 2026-10-30 |
| 77 | Gemini 3.1 Pro PreviewC | Google Gemini | C | 1.2 s | 1.3 s | 23% | 2026-10-30 |
| 78 | Gemini 3.1 Pro Preview Custom ToolsC | Google Gemini | C | 1.3 s | 1.8 s | 23% | 2026-05-31 |
| 79 | Claude Sonnet 5.5A | Anthropic | A | 1.3 s | 1.4 s | sin comparar | 2026-10-30 |
| 80 | gpt-5.5-2026-04-23A | OpenAI | A | 1.3 s | 2.1 s | 72% | 2026-10-30 |
| 81 | Claude Opus 5.5A | Anthropic | A | 1.3 s | 1.4 s | sin comparar | 2026-10-30 |
| 82 | Gemini Pro LatestC | Google Gemini | C | 1.4 s | 2.1 s | 19% | 2026-10-30 |
| 83 | Grok 4.3B | xAI | B | 1.4 s | 2.9 s | sin comparar | 2026-10-30 |
| 84 | Gemini 3.8 FlashA | Google Gemini | A | 1.5 s | 2.0 s | 38% | 2026-10-30 |
| 85 | Qwen3.7 MaxA | Alibaba Cloud Qwen (DashScope Intl) | A | 1.5 s | 2.3 s | sin comparar | 2026-10-30 |
| 86 | Grok Build 0.1B | xAI | B | 1.5 s | 1.6 s | sin comparar | 2026-10-30 |
| 87 | Gemini Flash LatestB | Google Gemini | B | 1.5 s | 1.7 s | 35% | 2026-10-30 |
| 88 | Qwen3.7 PlusB | Alibaba Cloud Qwen (DashScope Intl) | B | 1.5 s | 2.0 s | sin comparar | 2026-10-30 |
| 89 | Claude Opus 4.6B | Anthropic | B | 1.6 s | 1.7 s | 68% | 2026-10-30 |
| 90 | Gemini 3.7 FlashB | Google Gemini | B | 1.6 s | 2.2 s | 58% | 2026-10-30 |
| 91 | GLM-4.6V (vision)A | Z.ai (GLM / Zhipu) | A | 1.7 s | 3.1 s | 28% | 2026-10-30 |
| 92 | gpt-5-search-apiC | OpenAI | C | 1.7 s | 1.8 s | 62% | 2026-10-30 |
| 93 | gpt-5-search-api-2025-10-14B | OpenAI | B | 1.8 s | 1.8 s | 62% | 2026-10-30 |
| 94 | GLM-5.2A | Z.ai (GLM / Zhipu) | A | 2.0 s | 5.1 s | 28% | 2026-10-30 |
| 95 | GLM-4.5 AirB | Z.ai (GLM / Zhipu) | B | 2.1 s | 2.3 s | 35% | 2026-10-30 |
| 96 | Grok 4.6B | xAI | B | 2.4 s | 3.0 s | sin comparar | 2026-10-30 |
| 97 | gpt-oss-20bC | OVH AI Endpoints (GRA) | C | 2.4 s | 2.6 s | 35% | 2026-10-30 |
| 98 | GLM-5.1A | Z.ai (GLM / Zhipu) | A | 2.8 s | 6.9 s | 35% | 2026-10-30 |
| 99 | Claude Fable 5A | Anthropic | A | 2.8 s | 3.0 s | 62% | 2026-10-30 |
| 100 | GLM-5A | Z.ai (GLM / Zhipu) | A | 3.6 s | 6.3 s | 59% | 2026-10-30 |
| 101 | GLM-5 TurboB | Z.ai (GLM / Zhipu) | B | 5.0 s | 8.7 s | 33% | 2026-10-30 |
| 102 | GLM-4.6A | Z.ai (GLM / Zhipu) | A | 5.8 s | 7.3 s | 31% | 2026-10-30 |
| 103 | GLM-4.7A | Z.ai (GLM / Zhipu) | A | 6.4 s | 7.8 s | 60% | 2026-10-30 |
| 104 | GLM-4.5A | Z.ai (GLM / Zhipu) | A | 7.0 s | 7.2 s | 58% | 2026-10-30 |
| 105 | Qwen3.5-9BB | OVH AI Endpoints (GRA) | B | 9.7 s | 22.6 s | 23% | 2026-10-30 |
| · | Deep Research Max Preview (Apr-21-2026)B | Google Gemini | B | No se puede medir — responde por otra API | sin medir | sin comparar | — |
| · | Deep Research Preview (Apr-21-2026)B | Google Gemini | B | No se puede medir — responde por otra API | sin medir | sin comparar | — |
| · | Deep Research Pro Preview (Dec-12-2025)B | Google Gemini | B | No se puede medir — responde por otra API | sin medir | sin comparar | — |
| · | Gemini 2.5 Computer Use Preview 10-2025B | Google Gemini | B | No se puede medir — responde por otra API | sin medir | sin comparar | — |
| · | gpt-5.6-terra | Azure OpenAI (EU - Sweden) | sin nivel | No se puede medir — sin sonda para este proveedor | sin medir | 72% | — |
110 de 110 modelos · pulse una columna para ordenar
«No se puede medir» significa que el modelo no responde por la API de chat que cronometramos, o que su proveedor usa un protocolo que nuestra sonda todavía no habla. Esperar no lo arreglará.
Rápido (menos de 500 ms)
Medio (500–1000 ms)
Lento (más de 1000 ms)
Medido cuatro veces al día (02:00, 08:00, 14:00, 20:00 UTC) desde Ámsterdam.