τ³-Banking (AA) leaderboard
As of 6 Nov 2026, the top score on τ³-Banking (AA) is 51.3% by Qwen3.8 Max (Alibaba), ahead of Grok 4.6 at 50.7%, across 207 model configurations with a published result (Artificial Analysis, result dated 3 Aug 2026). The banking domain of τ²-bench: the model acts as a bank's support agent, follows policy, uses tools, and coordinates with a simulated customer who also acts. τ³-Banking (AA) counts toward the BenchLeader agents & tools score.
Key facts (as of 6 Nov 2026)
- Qwen3.8 Max holds the top τ³-Banking (AA) score at 51.3% (Artificial Analysis, 3 Aug 2026).
- Grok 4.6 is second on τ³-Banking (AA) at 50.7% (Artificial Analysis, 12 Aug 2026).
- τ³-Banking (AA): Artificial Analysis' τ²-Bench banking-domain run: a support agent using tools with a simulated customer (Artificial Analysis).
- 207 model configurations have a published τ³-Banking (AA) result on BenchLeader (BenchLeader, 6 Nov 2026).
- τ³-Banking (AA) feeds the agents & tools category of the BenchLeader Index (BenchLeader methodology).
Artificial Analysis' τ²-Bench banking-domain run: a support agent using tools with a simulated customer.
- Published by
- Artificial Analysis
- Category
- Agents & tools
- Index weight
- 1.0
- Models
- 207
- Data as of
- 6 Nov 2026
Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.
What the test looks like
The banking domain of τ²-bench: the model acts as a bank's support agent, follows policy, uses tools, and coordinates with a simulated customer who also acts.
How it is scored
Percent of conversations ending in the correct final state, run by Artificial Analysis.
What to keep in mind
One domain, with a simulated customer who is itself a model. Counts toward the agentic category since index v1.2, alongside the telecom run.
- 1Qwen3.8 Max (max)51.3%
- 2Grok 4.6 (high)50.7%
- 3Muse Spark 1.3 (max)50.5%
- 4GLM 5.3 (max)50.3%
- 5Qwen3.8 2.4T A95B49.1%
- 6Qwen3.8 27B (xhigh)48.0%
- 7Qwen3.8 27B (medium)47.4%
- 8Claude Fable 5.1 (max)47.2%
- 9Muse Spark 1.3 (xhigh)47.2%
- 10GLM 5.3 Flash47.2%
- 11Kimi K3 (max)46.0%
- 12Claude Fable 5.1 (xhigh)45.8%
- 13Gemini 3.8 Flash (medium)45.8%
- 14Qwen3.8-Flash-Next45.4%
- 15Gemini 3.8 Flash (high)45.0%
| # | ||||
|---|---|---|---|---|
| 1 |  Qwen3.8 Maxmax | 51.3% | 58.8 | 2026-08-03 |
| 2 |  Grok 4.6high | 50.7% | 60.5 | 2026-08-12 |
| 3 |  Muse Spark 1.3max | 50.5% | 62.0 | 2026-09-02 |
| 4 |  GLM 5.3maxopen ↗ | 50.3% | 59.0 | 2026-08-18 |
| 5 | Qwen3.8 2.4T A95Bopen ↗ | 49.1% | 52.0 | 2026-08-12 |
| 6 | Qwen3.8 27Bxhighopen ↗ | 48.0% | 52.5 | 2026-08-14 |
| 7 | Qwen3.8 27Bmediumopen ↗ | 47.4% | 50.3 | 2026-08-14 |
| 8 |  Claude Fable 5.1max | 47.2% | 67.2 | 2026-09-01 |
| 9 | Muse Spark 1.3xhigh | 47.2% | 58.5 | 2026-09-02 |
| 10 | GLM 5.3 Flashopen ↗ | 47.2% | 56.8 | 2026-08-26 |
| 11 |  Kimi K3maxopen ↗ | 46.0% | 62.1 | 2026-07-16 |
| 12 | Claude Fable 5.1xhigh | 45.8% | 65.4 | 2026-09-01 |
| 13 |  Gemini 3.8 Flashmedium | 45.8% | 60.7 | 2026-09-02 |
| 14 | Qwen3.8-Flash-Nextopen ↗ | 45.4% | 54.6 | 2026-08-26 |
| 15 | Gemini 3.8 Flashhigh | 45.0% | 62.0 | 2026-09-02 |
| 16 | Claude Opus 5high | 44.7% | 64.3 | 2026-07-24 |
| 17 |  GPT-5.6 Solmax | 44.3% | 65.4 | 2026-07-09 |
| 18 | Grok 4.6medium | 44.3% | 59.6 | 2026-08-12 |
| 19 | Claude Opus 5xhigh | 43.3% | 60.2 | 2026-07-24 |
| 20 | Grok 4.6xhigh | 43.3% | 57.9 | 2026-08-12 |
| 21 | GPT-6 Astraxhigh | 43.1% | 66.8 | 2026-09-03 |
| 22 | Claude Fable 5.1high | 43.1% | 64.0 | 2026-09-01 |
| 23 | Claude Opus 5max | 42.1% | 65.4 | 2026-07-24 |
| 24 | Grok 4.5high | 42.1% | 56.9 | 2026-07-08 |
| 25 | Kimi K3lowopen ↗ | 41.6% | 53.8 | 2026-07-16 |
| 26 | GPT-6 Astramax | 41.4% | 67.8 | 2026-09-03 |
| 27 | Claude Fable 5.1medium | 41.0% | 62.3 | 2026-09-01 |
| 28 |  Deepseek v4 Flash Visionmax | 41.0% | – | 2026-08-21 |
| 29 | NewInkling 2 Small | 40.4% | 57.4 | 2026-11-05 |
| 30 | GPT-5.6 Terramax | 40.2% | 60.7 | 2026-07-09 |
| 31 | GPT-6 Astrahigh | 40.0% | 67.0 | 2026-09-03 |
| 32 | GPT-5.4xhigh | 39.6% | 63.2 | 2026-03-05 |
| 33 | DeepSeek V4 Promaxopen ↗ | 39.6% | 57.0 | 2026-08-13 |
| 34 | DeepSeek V4 Flashmaxopen ↗ | 39.4% | 57.0 | 2026-07-31 |
| 35 | GPT-5.5xhigh | 39.0% | 64.3 | 2026-04-23 |
| 36 | Claude Fable 5.1low | 39.0% | 59.3 | 2026-09-01 |
| 37 | Claude Opus 5medium | 38.6% | 57.4 | 2026-07-24 |
| 38 |  Ling 3.0 Flash Finopen | 38.6% | – | 2026-09-11 |
| 39 | Claude Fable 5max | 38.1% | 65.3 | 2026-06-09 |
| 40 | GPT-5.6 Solxhigh | 38.1% | 64.2 | 2026-07-09 |
| 41 | Grok 4.6low | 38.1% | 55.2 | 2026-08-12 |
| 42 | Claude Sonnet 5max | 37.3% | 53.8 | 2026-06-30 |
| 43 | GPT-5.5high | 36.7% | 63.7 | 2026-04-23 |
| 44 | GPT-5.6 Solhigh | 36.7% | 63.6 | 2026-07-09 |
| 45 | GPT-5.6 Solmedium | 36.5% | 61.5 | 2026-07-09 |
| 46 |  Agnes 2.5 Pro Beta | 35.7% | – | 2026-08-26 |
| 47 | GPT-6 Astramedium | 35.5% | 65.1 | 2026-09-03 |
| 48 | Gemini 3.7 Flashmedium | 35.5% | 60.3 | 2026-08-13 |
| 49 |  Motif 3open | 35.3% | – | 2026-08-12 |
| 50 | Muse Spark 1.2xhigh | 34.9% | 61.7 | 2026-08-05 |
| 51 | GLM 5.2maxopen ↗ | 34.6% | 59.5 | 2026-06-16 |
| 52 | Claude Opus 4.7max | 34.6% | 59.5 | 2026-04-16 |
| 53 | Claude Sonnet 4.6max | 34.4% | 54.2 | 2026-02-17 |
| 54 | Ling 3.0 Flash VLopen ↗ | 34.4% | – | 2026-09-10 |
| 55 | Claude Opus 4.8max | 34.2% | 60.2 | 2026-05-28 |
| 56 |  K2 Horizon 375B A23Bopen | 34.2% | 49.8 | 2026-09-03 |
| 57 | Gemini 3.8 Flashlow | 33.2% | 56.2 | 2026-09-02 |
| 58 | Gemini 3.7 Flashhigh | 32.8% | 62.2 | 2026-08-13 |
| 59 | Gemini 3.5 Flashhigh | 32.2% | 62.5 | 2026-05-19 |
| 60 | Qwen3.8 27Blowopen ↗ | 32.2% | 49.1 | 2026-08-14 |
More agents & tools benchmarks · All LLM benchmarks
Related benchmarks
Top head-to-heads on τ³-Banking (AA)
Cite as: BenchLeader, “τ³-Banking (AA) leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/aa_tau_banking.
τ³-Banking (AA): questions
- What does τ³-Banking (AA) measure?
- The banking domain of τ²-bench: the model acts as a bank's support agent, follows policy, uses tools, and coordinates with a simulated customer who also acts. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on τ³-Banking (AA)?
- As of 6 Nov 2026, the highest score on τ³-Banking (AA) is 51.3% by Qwen3.8 Max (result dated 3 Aug 2026), ahead of Grok 4.6 at 50.7%. Scores come from Artificial Analysis and are re-read daily.
- Which AI model leads τ³-Banking (AA)?
- Qwen3.8 Max leads τ³-Banking (AA) with 51.3% as of 6 Nov 2026 (result dated 3 Aug 2026), ahead of Grok 4.6 at 50.7%.
- How many models have τ³-Banking (AA) results?
- 207 model configurations have a τ³-Banking (AA) result on BenchLeader, all taken from Artificial Analysis.
- Who runs τ³-Banking (AA) and how often is it updated?
- τ³-Banking (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does τ³-Banking (AA) count toward the BenchLeader Index?
- Yes. τ³-Banking (AA) contributes to the agents & tools category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.