Tau2 Airline: leaderboard
TAU2 airline domain benchmark for conversational agents in dual-control environments where both AI agents and users interact with tools in airline customer-service scenarios.
Metric: Pass@1 (%). Source: arxiv.org. Source: taubench.com. Source: benchmarklist.com. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 (High) | 84 | #83 (Claude Opus 4.5) |
| 2 | GPT-5.2 (High) | 83 | #110 (GPT-5.2) |
| 3 | GLM-5 (Thinking) | 82.5 | #141 (GLM-5) |
| 4 | Gemini 3 Flash (High) | 82.5 | #99 (Gemini 3 Flash) |
| 5 | Qwen 3.5 397B A17B (Reasoning) | 81.5 | #146 (Qwen 3.5 397B A17B) |
| 6 | Gemini 3 Pro (High) | 80.5 | #80 (Gemini 3 Pro) |
| 7 | Gemini 3 Pro | 73 | #80 |
| 8 | Claude Sonnet 4.5 (Thinking) | 72 | #142 (Claude Sonnet 4.5) |
| 9 | Claude Sonnet 4.5 | 70 | #142 |
| 10 | Qwen 3 Max (Thinking) | 69 | #208 (Qwen 3 Max) |
| 11 | Claude 3.7 Sonnet | 64.2 | #249 |
| 12 | DeepSeek V3.2 | 63.8 | #205 |
| 13 | GPT-5 | 62.5 | #98 |
| 14 | Claude Sonnet 4 | 60 | #199 |
| 15 | Claude Opus 4 | 59.6 | #163 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tau2-airline · How It Works · Data refreshed daily, snapshot 2026-10-20.