Tau3-Bench Banking_Knowledge: leaderboard
τ³-bench banking knowledge domain: RAG-based agentic tasks requiring document retrieval and reasoning. New domain in τ³-bench. From Sierra AI.
Metric: Pass@1 (%). Source: taubench.com. Saturation forecast: Around April 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.8 Max | 55.2 | #35 |
| 2 | Claude Opus 5 | 48.7 | #12 |
| 3 | Grok 4.5 | 47.9 | #48 |
| 4 | GPT-5.6 Sol | 46.9 | #19 |
| 5 | GPT-5.5 | 44.6 | #31 |
| 6 | Muse Spark 1.1 | 40.5 | #52 |
| 7 | Claude Opus 4.7 | 40.2 | #49 |
| 8 | Claude Opus 4.8 | 39.7 | #37 |
| 9 | Claude Fable 5 | 39.7 | #16 |
| 10 | GPT-5.4 | 39.4 | #79 |
| 11 | GLM-5.2 | 37.1 | #73 |
| 12 | Kimi K3 | 37.1 | #30 |
| 13 | GPT-5.2 | 32.2 | #110 |
| 14 | Claude Opus 4.6 | 27.3 | #63 |
| 15 | Gemini 3 Flash | 27.3 | #99 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tau3-bench-banking-knowledge · How It Works · Data refreshed daily, snapshot 2026-10-20.