τ³-Banking leaderboard
- Qwen3.8 Max (0902) leads
- 55.2%
- models measured
- 18
- lead over #2
- +6.4 pts
- median
- 39.6%
Leader on τ³-Banking · read November 5
On November 5, 2026, 18 of our catalog's models had a τ³-Banking result, and Qwen3.8 Max (0902) led with 55.2%, ahead of Anthropic Claude Opus 5 at 48.7%.
τ³-Banking is the knowledge-retrieval domain of Sierra's τ-bench. An agent talks with a simulated customer and must find and apply policy from a realistic knowledge base of about 700 documents, combining retrieval with policy rules. The number is Pass^1 as Sierra publishes it: the success rate of a single attempt.
How to read the score: We read the standard entries that use the AllTools retrieval setup, the one Sierra uses to standardize retrieval across models. Entries with a custom agent or another retriever are left out, so every number rests on the same setup.
Role in the index
τ³-Banking is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Qwen3.8 Max (0902)Qwen | 55.2% | 67.1 | $6.00 | |
| 2 | Claude Opus 5Anthropic | 48.7% | 81.6 | $25.00 | |
| 3 | Grok 4.5xAI | 47.9% | 52.3 | $6.00 | |
| 4 | GPT-5.6 SolOpenAI | 46.9% | 70.7 | $20.00 | |
| 5 | GPT-5.5OpenAI | 44.6% | 67.9 | $30.00 | |
| 6 | Muse Spark 1.1Meta | 40.5% | 70.3 | $4.25 | |
| 7 | Claude Opus 4.7Anthropic | 40.2% | 67.5 | $25.00 | |
| 8 | Claude Fable 5Anthropic | 39.7% | 75.9 | $50.00 | |
| 9 | Claude Opus 4.8Anthropic | 39.7% | 62.8 | $25.00 | |
| 10 | GPT-5.4OpenAI | 39.4% | 64.4 | $15.00 | |
| 11 | Kimi K3Moonshot AI | 37.1% | 62.3 | $15.00 | |
| 12 | GLM 5.2Z.ai | 37.1% | 61.1 | $4.40 | |
| 13 | GPT-5.2OpenAI | 32.2% | 55.9 | $14.00 | |
| 14 | Claude Opus 4.6Anthropic | 27.3% | 57.2 | $25.00 | |
| 15 | Gemini 3.1 Pro PreviewGoogle | 26.0% | 32.3 | $12.00 | |
| 16 | T InklingThinking Machines | 25.0% | 31.6 | $4.05 | |
| 17 | Claude Opus 4.5Anthropic | 24.7% | 44.0 | $25.00 | |
| 18 | Gemini 2.5 ProGoogle | 13.7% | 19.1 | $10.00 |
Only models in our catalog are listed, and each appears once with its best published run. 18 rows were read from the source on November 5, 2026, and rows that match no catalog model are left out.
The board above is tau-bench (Sierra)'s own run. Model makers also report τ³-Banking in their release announcements; those numbers are below, with the date and the announcement. They are not independent and do not count toward any index.
| Rank | Model | Score | Setting | Reported by | Date |
|---|---|---|---|---|---|
| 1 | Step 5 Preview | 42.5% | High | StepFun | October 8, 2026 (model release) |
| 2 | R Beam 1.0 501B A23B | 36.8% | FP8 | Reflection | October 16, 2026 (model release) |
| 3 | Kimi K3 | 33.4% | max | Moonshot AI | July 16, 2026 (model release) |
| 4 | Mercury 3 | 30.2% | MEDIUM | Inception | September 8, 2026 (model release) · read from the maker's chart |
| 5 | T Inkling | 23.7% | Not published | Thinking Machines | July 30, 2026 (model release) |
| 6 | Solar Pro 4 | 23.0% | Not published | Upstage | August 10, 2026 (model release) |
| 7 | T Inkling Small | 15.5% | Not published | Thinking Machines | July 30, 2026 (model release) |
| 8 | Nemotron 3 Super | 12.4% | Not published | NVIDIA | August 11, 2026 (model release) |
| 9 | Nemotron 3.5 Lightning | 9.3% | Not published | NVIDIA | August 11, 2026 (model release) |
| 10 | LFM2.5-24B-A2B | 8.0% | Not published | Liquid | October 13, 2026 (model release) |
| 11 | LFM2.5-2.6B | 5.7% | 2.6B | Liquid | August 11, 2026 (model release) |
Other reported runs
| Model | Score | Setting | Reported by | Date |
|---|---|---|---|---|
| R Beam 1.0 501B A23B | 36.1% | BF16 | Reflection | October 16, 2026 (model release) |
| R Beam 1.0 501B A23B | 35.4% | NVFP4 | Reflection | October 16, 2026 (model release) |