Modelos open-weight (Llama, Qwen, DeepSeek, Mistral…) classificados lado a lado com a fronteira proprietária (GPT, Claude, Gemini, Grok) — preços ao vivo, contexto, benchmarks e capacidades. Sem marketing — apenas os fatos.
| # | Model | Creator | Intelligence | Coding |
|---|---|---|---|---|
| 1 | Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) | Anthropic | 53.4 | 81.6 |
| 2 | GPT-6 Astra (max) | OpenAI | 52.8 | 76.9 |
| 3 | Claude Opus 5 (Adaptive Reasoning, Max Effort) | Anthropic | 50.7 | 78.0 |
| 4 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | Anthropic | 49.7 | 76.5 |
| 5 | Muse Spark 1.3 (max) | Meta | 48.2 | 75.8 |
| 6 | GPT-5.6 Sol (max) | OpenAI | 47.1 | 77.4 |
| 7 | GLM-5.3 (max) | Z AI | 44.9 | 74.8 |
| 8 | Grok 4.6 (high) | SpaceXAI | 44.4 | 76.8 |
| 9 | Kimi K3 (max) | Kimi | 43.8 | 76.2 |
| 10 | GPT-5.6 Terra (max) | OpenAI | 42.3 | 76.7 |
| 11 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | Anthropic | 42.0 | 74.3 |
| 12 | GLM-5.3-Flash | Z AI | 41.9 | 71.5 |
| 13 | Gemini 3.8 Flash (high) | 41.2 | 76.3 | |
| 14 | Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | Anthropic | 40.7 | 73.6 |
| 15 | Qwen3.8 Max | Alibaba | 40.3 | 71.8 |
| 16 | Qwen3.8 2.4T A95B | Alibaba | 40.0 | 71.9 |
| 17 | Qwen3.8-Flash-Next | Alibaba | 39.9 | 73.1 |
| 18 | Muse Spark 1.2 (xhigh) | Meta | 39.8 | 72.2 |
| 19 | Gemini 3.7 Flash (medium) | 39.6 | 71.5 | |
| 20 | DeepSeek V4.1 Flash (Reasoning, Max Effort) | DeepSeek | 39.5 | — |
| 21 | Grok 4.5 (high) | SpaceXAI | 39.1 | 72.4 |
| 22 | GPT-5.4 (xhigh) | OpenAI | 39.0 | 71.1 |
| 23 | GPT-5.5 (xhigh) | OpenAI | 38.6 | 74.9 |
| 24 | Claude Sonnet 5 (Adaptive Reasoning, Max Effort) | Anthropic | 38.4 | 71.5 |
| 25 | GPT-5.6 Luna (max) | OpenAI | 37.5 | 71.4 |
| 26 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) | DeepSeek | 36.3 | 68.8 |
| 27 | Agnes 3.0 Flash | Sapiens AI | 35.5 | — |
| 28 | Agnes 2.5 Pro Beta | Sapiens AI | 35.2 | 62.3 |
| 29 | DeepSeek V4 Flash Vision (Reasoning, Max Effort) | DeepSeek | 35.0 | 65.0 |
| 30 | DeepSeek V4 Flash 0731 (Reasoning, Max Effort) | DeepSeek | 34.5 | 69.1 |
| 31 | Gemini 3.6 Flash (high) | 34.3 | 69.2 | |
| 32 | Muse Spark 1.1 (xhigh) | Meta | 34.3 | 71.3 |
| 33 | GLM-5.2 (max) | Z AI | 34.0 | 68.8 |
| 34 | Qwen3.8 27B (xhigh) | Alibaba | 33.9 | 68.1 |
| 35 | Gemini 3.5 Flash (medium) | 33.6 | — | |
| 36 | Motif 3 | Motif Technologies | 33.6 | 63.5 |
| 37 | GPT-5.3 Codex (xhigh) | OpenAI | 32.5 | — |
| 38 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort) | Anthropic | 31.9 | — |
| 39 | Kimi K2.6 | Kimi | 31.3 | 61.8 |
| 40 | Muse Spark | Meta | 31.3 | 58.6 |
| 41 | DeepSeek V4 Pro (Reasoning, Max Effort) | DeepSeek | 30.9 | 59.4 |
| 42 | K2 Horizon 375B A23B | MBZUAI Institute of Foundation Models | 30.8 | 61.5 |
| 43 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) | Anthropic | 30.5 | 63.0 |
| 44 | Apodex 1.1 | Apodex | 30.4 | 60.8 |
| 45 | GPT-5.2 (xhigh) | OpenAI | 30.4 | — |
| 46 | Gemini 3.1 Pro Preview | 30.4 | 68.8 | |
| 47 | Qwen3.7 Max | Alibaba | 29.9 | 66.0 |
| 48 | MiniMax-M3 | MiniMax | 29.6 | 58.6 |
| 49 | Claude Opus 4.5 (Reasoning) | Anthropic | 29.1 | — |
| 50 | MiMo-V2-Pro | Xiaomi | 28.6 | — |
| 51 | GPT-5.2 Codex (xhigh) | OpenAI | 28.5 | — |
| 52 | Qwen3.6 Max Preview | Alibaba | 28.4 | — |
| 53 | Nex-N2-Pro | Nex AGI | 28.2 | 59.1 |
| 54 | Solar Pro 4 | Upstage | 28.2 | 52.7 |
| 55 | Gemini 3 Pro Preview (high) | 28.0 | — | |
| 56 | GLM-5 (Reasoning) | Z AI | 27.9 | — |
| 57 | Agnes 2.5 Pro Alpha | Sapiens AI | 27.8 | 58.8 |
| 58 | JT-4.1 Flash 236B A21B | China Mobile | 27.3 | 52.4 |
| 59 | Grok Build 0.1 0616 | SpaceXAI | 27.2 | 51.5 |
| 60 | Quasar 438B (max, based on GLM-5.2) | Multiverse Computing | 27.1 | 61.2 |
Como isso é classificado: modelos são ordenados pela popularidade de uso real (tokens semanais processados em aplicativos), via API OpenRouter. Apenas open-weight modelos estão incluídos (Llama, DeepSeek, Qwen, Mistral, Gemma, Kimi, GLM, Phi, Nemotron, e mais). O preço é por milhão de tokens; o contexto é a janela máxima. As setas de movimento comparam a classificação de hoje com a captura anterior. Os dados são atualizados diariamente.