OpenRouter Tau2-Bench Airline: leaderboard
OpenRouter's own τ²-Bench Airline run: 50 multi-turn airline customer-service tasks where the model must use tools and follow policy, executed against the production endpoints it serves.
Metric: Accuracy (%). Source: openrouter.ai. Status: saturation imminent. 122 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.7 Flash | 80.6 |
| 2 | Claude Fable 5 | 80 |
| 3 | Claude Opus 5 | 79.2 |
| 4 | Nova Micro | 78.7 |
| 5 | Qwen 3.5 397B A17B | 78.5 |
| 6 | Claude Fable 5.1 | 78.3 |
| 7 | Qwen 3.5 122B A10B | 77.3 |
| 8 | Gemini 3 Flash (Preview) | 77.3 |
| 9 | Step 3.7 Flash | 77.3 |
| 10 | DeepSeek V4 Pro (0813) | 77.2 |
| 11 | Claude Opus 4.5 | 77.1 |
| 12 | GLM-5.3 | 77.1 |
| 13 | Nemotron 3 Ultra | 76.9 |
| 14 | Qwen 3.8 27B | 76.8 |
| 15 | GPT-5.6 Sol | 76.7 |
Interactive version: theaggregate.ai/benchmark?slug=openrouter-tau2-bench-airline · How It Works · Data refreshed daily, snapshot 2026-09-20.