T1-Bench - Pass^3: leaderboard
T1-Bench (2026) tests tool-using assistants in multi-turn customer-service conversations across 11 service domains and 14 multi-domain combinations, with a simulated user following a written policy and three trials per conversation. This board is Pass^3, the percentage of conversations in which all three trials reproduce every required tool call, parameter and tool output; higher is better.
Metric: Pass^3 (%): share of conversations solved in all three trials, where a conversation passes when every ground-truth tool call and parameter is reproduced and every ground-truth tool output is matched; T1-Bench: 525 multi-turn customer-service conversations over 25 domain settings (11 single domains and 14 combinations of 2 to 11 domains, 76 tools), a Gemma4-26B-A4B-it user simulator, three trials per conversation at temperature 1.0; averages weighted by conversations per setting; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 31B (IT) | 41.14 | #342 |
| 2 | Gemma 4 26B A4B (IT) | 37.71 | #549 |
| 3 | gemma-4-E4B-it | 35.62 | #1469 |
| 4 | Claude Opus 4.6 | 34.48 | #102 |
| 5 | Claude Haiku 4.5 | 34.29 | #568 |
| 6 | Claude Sonnet 4.5 | 33.71 | #250 |
| 7 | GPT-OSS-120B (High) | 29.33 | #700 (GPT-OSS-120B) |
| 8 | Gemma 4 E2B (IT) | 28.38 | #2303 |
| 9 | GPT-5.4 (Low) | 20.95 | #123 (GPT-5.4) |
| 10 | GPT-5.4 Mini (Low) | 14.1 | #389 (GPT-5.4 Mini) |
| 11 | GPT-OSS-20B (High) | 8.57 | #1147 (GPT-OSS-20B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=t1-bench-pass-pow-3 · How It Works · Data refreshed daily, snapshot 2026-11-13.