总排行榜
智能、编程与智能体指数
综合指数来自 Artificial Analysis · 每 1M token 价格（输入 / 输出）
| # | 模型 | 智能 | 编程 | 智能体 | $ / 1M 输入 | $ / 1M 输出 |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) | 56.8 | 81.6 | 58.2 | $5.00 | $25.00 |
| 2 | GPT-6 Astra (max) | 54.7 | 76.9 | 51.6 | $11.00 | $55.00 |
| 3 | Claude Opus 5 (Adaptive Reasoning, Max Effort) | 54.1 | 78 | 56.4 | $10.00 | $50.00 |
| 4 | Qwen3.8 Max | 53.4 | 68.9 | 49.9 | — | — |
| 5 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | 53.2 | 76.5 | 51.2 | $10.00 | $50.00 |
| 6 | GPT-5.6 Sol (max) | 51.3 | 77.4 | 50.7 | $4.00 | $20.00 |
| 7 | Grok 4.6 (high) | 50.6 | 76.8 | 53.6 | $4.00 | $12.00 |
| 8 | Kimi K3 (max) | 50.2 | 76.2 | 50.9 | $3.00 | $15.00 |
| 9 | GLM-5.3 (max) | 48.6 | 74.8 | 53.6 | $1.40 | $4.40 |
| 10 | Gemini 3.8 Flash (high) | 47.1 | 76.3 | 41.2 | $1.35 | $6.75 |
| 11 | Qwen3.8 Max | 46.9 | 71.8 | 49.9 | $2.00 | $6.00 |
| 12 | GPT-5.6 Terra (max) | 46.8 | 76.7 | 43.9 | $4.00 | $24.00 |
| 13 | Muse Spark 1.2 (xhigh) | 46.8 | 72.2 | 44.2 | $1.25 | $4.25 |
| 14 | Qwen3.8 2.4T A95B | 46.7 | 71.9 | 50.7 | $2.00 | $6.00 |
| 15 | GLM-5.3-Flash | 46.2 | 71.5 | 51.5 | $0.150 | $0.500 |
| 16 | Grok 4.5 (high) | 45.5 | 72.4 | 42.4 | $4.00 | $12.00 |
| 17 | Gemini 3.7 Flash (high) | 45.2 | 76.1 | 36.6 | $1.35 | $6.75 |
| 18 | Claude Sonnet 5 (Adaptive Reasoning, Max Effort) | 45.1 | 71.5 | 44.5 | $2.00 | $10.00 |
| 19 | GPT-5.6 Luna (max) | 43.4 | 71.4 | 42.9 | $0.400 | $2.40 |
| 20 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort) | 42.1 | 68.8 | 42.5 | $1.10 | $3.30 |
| 21 | Qwen3.8 27B (xhigh) | 41.4 | 68.1 | 46.8 | $0.150 | $2.00 |
| 22 | DeepSeek V4 Flash 0731 (Reasoning, Max Effort) | 40.8 | 69.1 | 41.9 | $0.140 | $0.280 |
| 23 | Gemini 3.6 Flash (high) | 40.3 | 69.2 | 30.3 | $1.35 | $6.75 |
| 24 | Gemini 3.1 Pro Preview | 36.7 | 68.8 | 10.4 | $3.60 | $21.60 |
| 25 | MiniMax-M3 | 35.7 | 58.6 | 31 | $0 | $0 |
| 26 | MiMo-V2.5-Pro | 32.6 | 60.2 | 22.7 | $0.520 | $1.04 |
| 27 | Inkling (xhigh) | 32.2 | 52.1 | 24.4 | $1.00 | $4.05 |
| 28 | Gemini 3.5 Flash-Lite | 27.6 | 49.3 | 16.1 | $0.540 | $4.50 |
| 29 | Ling 3.0 Flash | 27.4 | 50.6 | 21.1 | $0.020 | $0.060 |
| 30 | Claude 4.5 Haiku (Reasoning) | 22.5 | 43.9 | 10.4 | $1.10 | $5.50 |
持续运行的基准测试
推理、智能体与搜索：在实时模型上反复运行
GPQA Diamond
研究生水平的科学推理
| # | 模型 | 得分 | $ / 任务 |
|---|---|---|---|
| 1 | Gemini 3.1 Pro Preview | 94.4% | $0.202 |
| 2 | GPT-6 Astra | 94.4% | $0.099 |
| 3 | Gemini 3.7 Flash | 94.3% | $0.031 |
| 4 | GPT-5.5 | 93.8% | $0.308 |
| 5 | GPT-5.6 Sol Pro | 93.8% | $0.445 |
| 6 | Grok 4.6 | 93.3% | $0.122 |
| 7 | Gemini 3.6 Flash | 92.8% | $0.067 |
| 8 | Gemini 3.5 Flash | 92.8% | $0.142 |
| 9 | GPT-5.6 Sol | 91.9% | $0.071 |
| 10 | Kimi K3 | 91.5% | $0.147 |
| 11 | Claude Fable 5.1 | 90.9% | $0.123 |
| 12 | MiniMax M3 | 90.5% | $0.030 |
| 13 | GPT-5.4 | 90.3% | $0.147 |
| 14 | GPT-5.6 Luna Pro | 90.3% | $0.044 |
| 15 | Claude Opus 4.8 | 89.4% | $0.198 |
为什么这很重要
没有哪个模型能赢下所有基准测试。
领先者因任务而异：一个模型在推理上夺冠，另一个在编程上领先，第三个在智能体搜索中胜出。这正是 MultipleChat 存在的意义：把领先模型并排放在一起，让它们起草、互相质疑和验证，最终留下最好的答案。
质量、成本和速度兼顾
不标价格的基准分数只是营销。这里的每个排行榜都把质量与单任务成本或每百万 token 成本放在一起。
实时且开放
持续更新，并可下载为 JSON。我们自己的季度对比研究请见 MultipleChat AI 基准测试。
署名
来源：OpenRouter（openrouter.ai/rankings），含 Artificial Analysis 指数，数据截至 2026-09-07T14:28:14Z。依据 CC BY 4.0 许可发布。