OpenRouter · 客服智能体（τ-bench 航空）
定期获取并展示公开的基准测试资料（LMArena 用户投票排名、Epoch AI 测试分数、OpenRouter 使用量排名）。每个榜单均标注数据发布日期和获取日期。这些分数并非本站自行评定。
如何解读此排行榜
衡量什么： 这是 OpenRouter 运行 τ-bench（验证版，航空领域）的结果。评测模型在航空公司客服场景中，能否使用工具按规则处理更改预订等请求。
如何解读分数： 成功率（%）越高越好。也可同时比较每题平均成本。
注意事项： 仅涵盖一个领域（航空）的场景，因此可能与其他业务中的智能体表现不同。
本次排名的解读要点
- 最高分为 Claude Fable 5 的 79.9%。
- 第 1 名与第 5 名的差距为 1.4%p。
- 前 10 个模型中，Anthropic 的模型有 4 个，数量最多。
- 前 10 个模型中，每题平均成本最低的是 Step 3.7 Flash（$0.0201，得分 77.3%）。
客服智能体（τ-bench 航空） OpenRouter
发布日期：2026-10-04 · 获取日期：2026-10-05
图表仅显示同一模型各推理设置（high、max 等）中排名最高的一项。表格列出所有设置各自的排名。
查看表格（前 60 名）
| 排名 | 模型 | 开发方 | 分数 | 每题平均成本 |
|---|---|---|---|---|
| 1 | #1Claude Fable 5 | Anthropic | 79.9% | $0.9265 |
| 2 | #2MiMo-V2.6-Flash | Xiaomi | 79.3% | $0.0204 |
| 3 | #3Claude Fable 5.1 | Anthropic | 79.2% | $0.7162 |
| 4 | #4Claude Opus 5 | Anthropic | 78.6% | $0.4868 |
| 5 | #5Gemini 3.7 Flash | 78.5% | $0.1149 | |
| 6 | #6ANova Micro 1.0 | Amazon | 78% | $1.2758 |
| 7 | #7SStep 3.7 Flash | StepFun | 77.3% | $0.0201 |
| 8 | #8Claude Opus 4.5 | Anthropic | 77.1% | $0.5319 |
| 9 | #9OGPT-5.1 | OpenAI | 77.1% | $0.4149 |
| 10 | #10Qwen3.8 27B | Alibaba | 77% | $0.0811 |
| 11 | #11ZGLM 5 | Z.ai | 77% | $0.0447 |
| 12 | #12DeepSeek V4 Pro 0813 | DeepSeek | 77% | $0.1001 |
| 13 | #13Qwen3.5 397B A17B | Alibaba | 76.9% | $0.1697 |
| 14 | #14Gemini 3.5 Flash Lite | 76.9% | $0.0978 | |
| 15 | #15Gemma 4 31B | 76.7% | $0.0282 | |
| 16 | #16Claude Opus 4.8 | Anthropic | 76.4% | $0.4977 |
| 17 | #17DeepSeek V4 Pro 0423 | DeepSeek | 76.4% | $0.0399 |
| 18 | #18DeepSeek V4.1 Flash | DeepSeek | 76.2% | $0.0315 |
| 19 | #19ZGLM 5.3 | Z.ai | 76.1% | $0.062 |
| 20 | #20Qwen3.5-122B-A10B | Alibaba | 76.1% | $0.169 |
| 21 | #21ZGLM 5.3 | Z.ai | 76% | $0.016 |
| 22 | #22Claude Opus 4.7 | Anthropic | 75.9% | $0.4192 |
| 23 | #23Gemini 3.1 Pro Preview | 75.8% | $0.3592 | |
| 24 | #24ZGLM 5.3 Flash | Z.ai | 75.6% | $0.0067 |
| 25 | #25Qwen3.8 2.4T A95B | Alibaba | 75.5% | $0.1978 |
| 26 | #26ZGLM 5.1 | Z.ai | 75.4% | $0.0778 |
| 27 | #27xGrok 4.6 | xAI | 75.3% | $0.3309 |
| 28 | #28Claude Sonnet 5 | Anthropic | 75.3% | $0.1994 |
| 29 | #29OGPT-5.4 | OpenAI | 75.3% | $0.2916 |
| 30 | #30Gemini 3 Flash Preview | 75.3% | $0.1221 | |
| 31 | #31OGPT-5.3-Codex | OpenAI | 75.3% | $0.2997 |
| 32 | #32OGPT-5.5 | OpenAI | 75.1% | $0.4875 |
| 33 | #33DeepSeek V4 Flash Vision Exp | DeepSeek | 74.8% | $0.0302 |
| 34 | #34Claude Opus 4.6 | Anthropic | 74.7% | $0.4898 |
| 35 | #35Claude Sonnet 5.5 | Anthropic | 74.7% | $0.1787 |
| 36 | #36Gemini 3.6 Flash | 74.7% | $0.2141 | |
| 37 | #37OGPT-5.6 Sol | OpenAI | 74.7% | $0.184 |
| 38 | #38Muse Glimmer 30B | Meta | 74.7% | $0.0374 |
| 39 | #39Claude Sonnet 4.6 | Anthropic | 74.6% | $0.3238 |
| 40 | #40DeepSeek V4 Flash 0423 | DeepSeek | 74.4% | $0.0111 |
| 41 | #41Claude Opus 5.5 | Anthropic | 74.3% | $0.3458 |
| 42 | #42Qwen3.6 27B | Alibaba | 74.3% | $0.1343 |
| 43 | #43Kimi K2.6 | Moonshot AI | 74.1% | $0.0733 |
| 44 | #44MiniMax M3 | MiniMax | 74.1% | $0.0235 |
| 45 | #45DeepSeek V3.2 | DeepSeek | 74% | $0.0299 |
| 46 | #46OGPT-5.6 Sol Pro | OpenAI | 73.7% | $1.288 |
| 47 | #47MiMo-V2.5-Pro | Xiaomi | 73.6% | $0.0304 |
| 48 | #48DeepSeek V4 Flash 0731 | DeepSeek | 73.4% | $0.0092 |
| 49 | #49OGPT-5.2 | OpenAI | 73.3% | $0.2413 |
| 50 | #50OGPT-5.6 Terra | OpenAI | 73.1% | $0.1054 |
| 51 | #51Gemini 3.5 Flash | 73.1% | $0.3988 | |
| 52 | #52OGPT-6 Astra Pro | OpenAI | 72.7% | $2.9136 |
| 53 | #53ZGLM 5.2 | Z.ai | 72.7% | $0.034 |
| 54 | #54Kimi K2.7 Code | Moonshot AI | 72.6% | $0.0736 |
| 55 | #55Claude Sonnet 4.5 | Anthropic | 72.4% | $0.3328 |
| 56 | #56ZGLM 4.6 | Z.ai | 72.1% | $0.0499 |
| 57 | #57ZGLM 4.7 | Z.ai | 72.1% | $0.0581 |
| 58 | #58Nemotron 3 Ultra | NVIDIA | 72% | $0.1515 |
| 59 | #59Qwen3.7 Max | Alibaba | 72% | $0.4029 |
| 60 | #60Gemini 3.1 Flash Lite | 72% | $0.0932 |
Source: OpenRouter evals (openrouter.ai) via OpenRouter (openrouter.ai/rankings). Licensed under CC BY 4.0.
数据：LMArena leaderboard dataset（CC BY 4.0）。修改：各榜单仅显示前 50 名，分数四舍五入。 https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset公司标志为各公司的商标，仅用于帮助区分（图标：Simple Icons）。数据：Epoch AI — Capabilities & benchmarking（CC BY 4.0）。 https://epoch.ai/benchmarksSource: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings). Source: OpenRouter evals (openrouter.ai) via OpenRouter (openrouter.ai/rankings). Source: Design Arena (www.designarena.ai) via OpenRouter (openrouter.ai/rankings). Licensed under CC BY 4.0. Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-05. Licensed under CC BY 4.0. https://openrouter.ai/rankings