Agentic · tools

τ³-bench results leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on τ³-bench results is 96.0% by Mercury 2.5. 74 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	DeepSeek V3.2	100.0%	estimated ± 8.9 pp, low confidence
2	GLM-5V-Turbo	100.0%	estimated ± 8.9 pp, low confidence
3	GPT-4.1	100.0%	estimated ± 8.9 pp, low confidence
4	GPT-OSS 120B	100.0%	estimated ± 8.9 pp, low confidence
5	Kimi K2.5 (Reasoning)	100.0%	estimated ± 8.9 pp, low confidence
6	Qwen3.5-35B-A3B	100.0%	estimated ± 8.9 pp, low confidence
7	Trinity-Large-Thinking	100.0%	estimated ± 8.9 pp, low confidence
8	Gemma 4 31B	100.0%	estimated ± 8.9 pp, low confidence
9	MiMo-V2-Pro	99.9%	estimated ± 8.9 pp, low confidence
10	Hy3 Preview	99.9%	estimated ± 8.9 pp, low confidence
11	Grok 4.20	97.5%	estimated ± 8.9 pp, low confidence
12	Gemini 3.1 Flash-Lite	97.0%	estimated ± 8.9 pp, low confidence
13	Mercury 2.5	96.0%	measured
14	Mistral Medium 3.5 128B	91.4%	measured
15	Qwen3.5-27B	87.1%	estimated ± 8.9 pp, medium confidence
16	Claude 4 Sonnet	83.2%	estimated ± 8.9 pp, medium confidence
17	GLM-4.7	78.9%	estimated ± 8.9 pp, medium confidence
18	MiMo-V2.5-Pro	72.9%	measured
19	GPT-5.6 Sol	72.9%	estimated ± 2.1 pp, low confidence
20	GPT-5.6 Luna	71.9%	estimated ± 2.1 pp, low confidence
21	GPT-5.6 Terra	71.8%	estimated ± 2.1 pp, low confidence
22	Step 3.7 Flash	71.0%	estimated ± 2.1 pp, low confidence
23	Nemotron 3 Ultra	70.9%	measured
24	Qwen3.6 Plus	70.7%	measured
25	Qwen3.7 Plus	70.7%	estimated ± 1.4 pp, low confidence
26	GLM-5.1	70.6%	measured
27	MiniMax M2.7	70.3%	estimated ± 2.1 pp, low confidence
28	Claude Opus 4.5	70.2%	measured
29	GPT-5.1	70.0%	estimated ± 8.9 pp, medium confidence
30	Muse Spark 1.1	69.6%	estimated ± 1.8 pp, low confidence
31	Claude Opus 5	69.6%	estimated ± 1.8 pp, low confidence
32	Step 5 Preview	69.6%	estimated ± 1.8 pp, low confidence
33	Kimi K3	69.6%	estimated ± 1.8 pp, low confidence
34	Gemini 3.5 Flash	69.6%	estimated ± 1.8 pp, low confidence
35	Hy4 preview	69.6%	estimated ± 1.8 pp, low confidence
36	Claude Opus 4.8	69.6%	estimated ± 1.8 pp, low confidence
37	Ornith-1.5-397B	69.6%	estimated ± 1.8 pp, low confidence
38	Inkling-Small	69.6%	estimated ± 1.8 pp, low confidence
39	Beam	69.6%	estimated ± 1.8 pp, low confidence
40	Claude Opus 4.7 (Adaptive)	69.6%	estimated ± 1.8 pp, low confidence
41	GLM-5.2	69.6%	estimated ± 1.8 pp, low confidence
42	Qwen3.7 Max	69.6%	estimated ± 1.8 pp, low confidence
43	Kimi K2.7 Code	69.6%	estimated ± 1.8 pp, low confidence
44	Muse Glimmer 30B	69.6%	estimated ± 1.8 pp, low confidence
45	GPT-5.5	69.6%	estimated ± 1.8 pp, low confidence
46	MiniMax M3	69.6%	estimated ± 1.8 pp, low confidence
47	Inkling	69.6%	estimated ± 1.8 pp, low confidence
48	DeepSeek V4 Pro 0813	69.6%	estimated ± 1.8 pp, low confidence
49	GPT-5.4	69.6%	estimated ± 1.8 pp, medium confidence
50	Ornith-1.5-35B-A3B	69.6%	estimated ± 1.8 pp, medium confidence
51	DeepSeek V4 Flash 0731	69.6%	estimated ± 1.8 pp, medium confidence
52	Ling 3.0 Flash	69.6%	estimated ± 1.8 pp, medium confidence
53	Solar Pro 4	69.6%	estimated ± 1.8 pp, medium confidence
54	Solar Open 2	69.5%	estimated ± 1.8 pp, medium confidence
55	GPT-5.4 mini	69.5%	estimated ± 1.8 pp, medium confidence
56	GPT-5.4 nano	69.5%	estimated ± 1.8 pp, medium confidence
57	Kimi K2.6	69.5%	estimated ± 1.8 pp, medium confidence
58	Ornith-1.5-9B	69.5%	estimated ± 1.8 pp, medium confidence
59	LLaDA2.2-flash	69.3%	estimated ± 1.8 pp, medium confidence
60	LongCat-Flash-Lite-Sparse	69.2%	estimated ± 1.8 pp, medium confidence
61	Gemini 2.5 Pro	69.1%	estimated ± 8.9 pp, medium confidence
62	Grok 4	69.0%	estimated ± 8.9 pp, medium confidence
63	Qwen3 Max	68.9%	estimated ± 8.9 pp, medium confidence
64	Grok 4.3	68.9%	estimated ± 8.9 pp, medium confidence
65	Claude Opus 4.6	68.9%	estimated ± 8.9 pp, medium confidence
66	Claude Opus 4.7	68.9%	estimated ± 8.9 pp, low confidence
67	Claude Sonnet 4.5	68.9%	estimated ± 8.9 pp, medium confidence
68	Claude Sonnet 4.6	68.9%	estimated ± 8.9 pp, medium confidence
69	Gemini 3.1 Pro	68.9%	estimated ± 8.9 pp, medium confidence
70	Gemini 3 Flash	68.9%	estimated ± 8.9 pp, medium confidence
71	Gemini 3 Pro	68.9%	estimated ± 8.9 pp, medium confidence
72	GPT-5.1-Codex	68.9%	estimated ± 8.9 pp, medium confidence
73	GPT-5.2	68.9%	estimated ± 8.9 pp, medium confidence
74	GPT-5.2-Codex	68.9%	estimated ± 8.9 pp, medium confidence
75	GPT-5.3 Codex	68.9%	estimated ± 8.9 pp, medium confidence
76	Grok 4.1 Fast	68.9%	estimated ± 8.9 pp, medium confidence
77	Grok Build 0.1	68.9%	estimated ± 8.9 pp, medium confidence
78	MiMo-V2.5	68.9%	estimated ± 8.9 pp, medium confidence
79	Qwen3.6-27B	68.9%	estimated ± 8.9 pp, medium confidence
80	Qwen3.5 397B	68.4%	measured
81	dots3-note Preview	67.9%	estimated ± 11.4 pp, low confidence
82	Qwen3.6-35B-A3B	67.2%	measured
83	Pokee-Isaac 28B	66.2%	measured
84	Kimi K2.5	65.7%	measured
85	GLM-5	65.6%	measured
86	Granite 4.2 30B	62.0%	measured
87	Granite 4.2 8B	58.1%	measured
88	Granite 4.2 3B	45.8%	measured
89	Qwen3.8 Max	41.2%	estimated ± 11.4 pp, low confidence
90	Atria Dawn Preview	41.2%	measured
91	Nemotron 3.5 Lightning 30B A3B NVFP4	9.5%	measured
92	LFM2.5-2.6B	5.7%	measured

τ³-bench results leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General