Benchmarks · agentic

τ³-Banking

τ³-Bench banking domain: multi-turn tool-using agent tasks with a simulated user; pass rate.

99 models · 153 results · data as of 2026-11-06 · artificialanalysis.ai ↗

More from this benchmark

4 published measures

Score vs any published measure4 measures: sub-scores, splits, counts

Highlight models

Click a point on the chart to highlight that exact effort; click it again to remove it.

Chart settings

99 of 99 selected

73 more models under All models in settings.

darker is better within each column · click a column to sort

153 rows

#	Model	Effort	τ³-Banking
1	Qwen 3.8 Max	max	51.3%
2	Grok 4.6	high	50.7%
3	Muse Spark 1.3	max	50.5%
4	GLM 5.3	max	50.3%
5	Qwen 3.8 2.4T-A95B		49.1%
6	Qwen 3.8 27B	xhigh	48.0%
7	Qwen 3.8 27B	medium	47.4%
8	Fable 5.1	max	47.2%
9	GLM 5.3 Flash	max	47.2%
10	Muse Spark 1.3	xhigh	47.2%
11	Kimi K3	max	46.0%
12	Fable 5.1	xhigh	45.8%
13	Gemini 3.8 Flash	medium	45.8%
14	Qwen 3.8 Flash		45.4%
15	Gemini 3.8 Flash	high	44.9%
16	Opus 5	high	44.7%
17	GPT-5.6 Sol	max	44.3%
18	Grok 4.6	medium	44.3%
19	Opus 5	xhigh	43.3%
20	Grok 4.6	xhigh	43.3%
21	Fable 5.1	high	43.1%
22	GPT-6 Astra	xhigh	43.1%
23	Opus 5	max	42.1%
24	Grok 4.5	high	42.1%
25	Kimi K3	low	41.6%
26	GPT-6 Astra	max	41.4%
27	Fable 5.1	medium	41.0%
28	DeepSeek V4 Flash	max	41.0%
29	Inkling 2 Small		40.4%
30	GPT-5.6 Terra	max	40.2%