综合排行榜
智能、编程与智能体指数
Artificial Analysis 综合指数 · 每百万 token 价格（输入/输出）
| # | 模型 | 智能 | 编程 | 智能体 | $ / 1M 输入 | $ / 1M 输出 |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (Max, Default Fallback) | 58.4 | — | — | $4.40 | $22.00 |
| 2 | GPT-6 Astra (Max) | 56.9 | 76.9 | 51 | $10.00 | $50.00 |
| 3 | GPT-6.1 Sol (Max) | 54.5 | — | — | $2.00 | $10.00 |
| 4 | Claude Sonnet 5.5 (Max, Default Fallback) | 54.2 | — | — | $2.20 | $11.00 |
| 5 | Qwen3.8 Max | 53.4 | 68.9 | 49.9 | — | — |
| 6 | GPT-6.1 Terra (Max) | 51 | — | — | $1.20 | $6.00 |
| 7 | Gemini 4 Argon (High) | 50.4 | — | — | $2.20 | $11.00 |
| 8 | Claude Fable 5.1 (Xhigh, Default Fallback) | 49.7 | 80.7 | 57.2 | $5.00 | $25.00 |
| 9 | Kimi K3.1 Pro | 46 | — | — | $4.00 | $20.00 |
| 10 | Muse Wonder (Max) | 43.3 | — | — | $4.00 | $20.00 |
| 11 | Muse Spark 1.3 (Max) | 36.4 | 75.8 | 55.5 | $1.25 | $4.25 |
| 12 | Claude Haiku 5.5 (Max) | 36.2 | — | — | $0.110 | $0.550 |
| 13 | Grok 4.7 (High) | 35 | — | — | $1.60 | $4.80 |
| 14 | Qwen3.8 Max (0902) | 33.6 | 76.2 | 56 | $2.00 | $6.00 |
| 15 | GLM-5.3 (Max) | 31.7 | 74.8 | 53.1 | $0.980 | $3.08 |
| 16 | Gemini 3.8 Flash (High) | 30.5 | 76.3 | 40.2 | $0.830 | $4.12 |
| 17 | MiMo-V2.6-Pro | 29.9 | — | — | $0.520 | $1.04 |
| 18 | MiniMax-M3.1-Flash | 28.9 | — | — | $0.140 | $0.540 |
| 19 | DeepSeek V4.1 Flash 1017 (High) | 28.5 | — | — | $0.020 | $0.250 |
| 20 | Step 5 Preview | 28.5 | — | — | $1.00 | $2.70 |
| 21 | Solar Pro 4 | 28.2 | 52.7 | — | $0.150 | $0.600 |
| 22 | GLM-5.3-Flash | 27.8 | 71.5 | 50.9 | $0.070 | $0.250 |
| 23 | GPT-6 Luna (Max) | 27.8 | — | — | $0.100 | $0.500 |
| 24 | MiMo-V2.6-Flash | 26.7 | — | — | $0.140 | $0.280 |
| 25 | Mistral Large 4 (High) | 26 | — | — | $0.770 | $2.31 |
| 26 | Muse Spark Open (Xhigh) | 24.7 | — | — | $1.25 | $4.25 |
| 27 | DeepSeek V4 Pro 0813 (Max) | 23.1 | 68.8 | 41.3 | $0.110 | $4.00 |
| 28 | Qwen3.8 27B (Xhigh) | 20.1 | 68.1 | 45.8 | $0.300 | $1.80 |
| 29 | MiniMax-M3 | 18.7 | 58.6 | 29.5 | $0.220 | $0.880 |
| 30 | Beam 1.0 | 17.7 | — | — | $0.400 | $1.20 |
持续运行的基准测试
推理、智能体与搜索：在最新模型上重新运行
GPQA Diamond
研究生水平的科学推理
| # | 模型 | 得分 | $ / 任务 |
|---|---|---|---|
| 1 | Gemini 4 Argon | 96.5% | $0.095 |
| 2 | Gemini 4 Argon | 96.0% | $0.029 |
| 3 | Gemini 3.8 Flash | 95.6% | $0.068 |
| 4 | GPT-6 Astra Pro | 95.5% | $0.333 |
| 5 | Fugu Ultra | 94.6% | $1.15 |
| 6 | Gemini 3.1 Pro Preview | 94.5% | $0.202 |
| 7 | GPT-6 Astra | 94.4% | $0.113 |
| 8 | GPT-6.1 Sol | 94.4% | $0.021 |
| 9 | GPT-5.6 Sol Pro | 94.1% | $0.151 |
| 10 | GPT-5.5 | 93.9% | $0.313 |
| 11 | Gemini 3.7 Flash | 93.9% | $0.029 |
| 12 | Gemini 3.8 Flash | 93.4% | $0.023 |
| 13 | Fugu Ultra V2 | 93.3% | $0.503 |
| 14 | Jev Router | 93.3% | $0.035 |
| 15 | Grok 4.6 | 92.8% | $0.123 |
为什么重要
没有哪个模型能在所有基准测试中胜出。
领先者随任务而变：推理是这个模型最强，编程是那个，智能体搜索又是另一个。这正是 MultipleChat 存在的理由：把领先模型并排放在一起，让它们起草、相互质疑、彼此核验，你只取最好的答案。
质量、成本、速度一起看
不标价格的基准测试分数只是营销。这里的每个排行榜都把质量与每任务、每百万 token 的成本放在一起展示。
实时且开放
持续更新，可下载为 JSON 格式。我们自己的季度对比研究请见 MultipleChat AI Benchmark。
来源
Source: OpenRouter (openrouter.ai/rankings)（含 Artificial Analysis 指数）, as of 2026-11-03T23:35:37Z。许可协议：CC BY 4.0。