Tau-Bench Telecom: leaderboard
Agentic customer service benchmark in a simulated telecom environment. Models handle complex multi-turn conversations with tool-calling. From Sierra AI (tau2-bench).
Metric: Pass@1 (%). Source: taubench.com. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 Max (Thinking) | 98.2 | #208 (Qwen 3 Max) |
| 2 | Claude Sonnet 4.5 | 98 | #142 |
| 3 | Gemini 3 Pro | 98 | #80 |
| 4 | DeepSeek V3.2 | 96.2 | #205 |
| 5 | GPT-5 | 95.8 | #98 |
| 6 | Qwen 3 Max | 84.2 | #208 |
| 7 | Kimi K2 | 65.8 | #243 |
| 8 | O4 Mini | 50.2 | #180 |
| 9 | Claude 3.7 Sonnet | 49 | #249 |
| 10 | GPT-4.1 Mini | 48.9 | #355 |
| 11 | GPT-4.1 | 34 | #248 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tau-bench-telecom · How It Works · Data refreshed daily, snapshot 2026-10-20.