Agentic · terminal

Terminal-Bench Hard leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on Terminal-Bench Hard is 65.9% by GPT-5.6 Sol. 64 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	68.3%	estimated ± 3.8 pp, medium confidence
2	Claude Fable 5.1	66.3%	estimated ± 3.8 pp, high confidence
3	Claude Opus 5	65.9%	estimated ± 3.8 pp, high confidence
4	GPT-5.6 Sol	65.9%	measured
5	Gemini 3.8 Flash	63.0%	estimated ± 3.8 pp, high confidence
6	Claude Fable 5	62.9%	measured
7	Kimi K3	62.7%	estimated ± 3.8 pp, high confidence
8	GPT-5.6 Luna	61.0%	estimated ± 3.8 pp, high confidence
9	GPT-5.5	60.6%	measured
10	Grok 4.6	60.4%	estimated ± 3.8 pp, high confidence
11	Gemini 3.7 Flash	59.7%	estimated ± 3.8 pp, high confidence
12	GPT-5.6 Terra	59.7%	estimated ± 3.8 pp, high confidence
13	Claude Opus 4.8	58.3%	measured
14	GPT-5.4	57.6%	measured
15	Claude Sonnet 5	57.2%	estimated ± 3.8 pp, high confidence
16	DeepSeek V4.1 Flash 1017	57.2%	estimated ± 3.8 pp, high confidence
17	Gemini 3.5 Flash	57.0%	estimated ± 3.8 pp, high confidence
18	Gemini 3.6 Flash	56.6%	estimated ± 3.8 pp, high confidence
19	Grok 4.7	56.3%	estimated ± 3.8 pp, high confidence
20	Muse Spark 1.3	55.5%	estimated ± 3.8 pp, high confidence
21	GPT-5.3 Codex	55.3%	estimated ± 4.0 pp, high confidence
22	GLM-5.3	54.9%	estimated ± 3.8 pp, high confidence
23	Claude Opus 4.7	54.5%	measured
24	Gemini 3.1 Pro	53.8%	measured
25	Muse Spark 1.2	53.5%	estimated ± 3.8 pp, high confidence
26	Muse Spark 1.1	53.3%	estimated ± 3.8 pp, high confidence
27	GPT-5.4 mini	52.3%	measured
28	Grok 4.5	52.2%	estimated ± 3.8 pp, high confidence
29	Qwen3.8 Max	52.0%	estimated ± 3.8 pp, high confidence
30	DeepSeek V4 Flash 0731	51.7%	estimated ± 3.8 pp, high confidence
31	GLM-5.2	50.8%	measured
32	Qwen3.7 Max	50.8%	measured
33	GLM-5.3-Flash	49.4%	estimated ± 3.8 pp, high confidence
34	Claude Opus 4.7 (Adaptive)	49.2%	estimated ± 4.0 pp, high confidence
35	Claude Opus 4.6	48.5%	measured
36	MiMo-V2.5	48.3%	estimated ± 3.8 pp, high confidence
37	Composer 2.5	47.3%	estimated ± 3.8 pp, high confidence
38	Qwen3.8-27B	47.3%	estimated ± 3.8 pp, high confidence
39	Claude Opus 4.5 Thinking	47.0%	measured
40	GPT-5.2	47.0%	measured
41	Qwen3.7 Plus	47.0%	measured
42	Claude Sonnet 4.6	46.9%	estimated ± 3.8 pp, high confidence
43	GLM-5.1	46.8%	estimated ± 3.8 pp, high confidence
44	Hy4 preview	46.2%	estimated ± 3.8 pp, high confidence
45	Inkling-Small	46.2%	estimated ± 3.8 pp, high confidence
46	DeepSeek V4 Pro 0813	46.1%	estimated ± 3.8 pp, high confidence
47	Gemini 3 Flash	45.8%	estimated ± 3.8 pp, high confidence
48	MiniMax M3	45.7%	estimated ± 3.8 pp, high confidence
49	GPT-5.1	45.5%	measured
50	Muse Spark	45.5%	measured
51	Composer 2	45.3%	estimated ± 4.0 pp, high confidence
52	Nemotron 3 Ultra	45.1%	estimated ± 3.8 pp, medium confidence
53	Gemini 3.5 Flash-Lite	44.9%	estimated ± 3.8 pp, medium confidence
54	Ling 3.0 Flash	44.9%	estimated ± 3.8 pp, medium confidence
55	Kimi K2.7 Code	44.7%	measured
56	MiniMax M2.7	44.7%	estimated ± 3.8 pp, medium confidence
57	Inkling	44.5%	estimated ± 3.8 pp, medium confidence
58	Claude Opus 4.5	44.4%	estimated ± 4.0 pp, high confidence
59	Qwen3.6-27B	44.4%	estimated ± 4.0 pp, high confidence
60	Grok 4.20	44.0%	estimated ± 3.8 pp, medium confidence
61	Claude Haiku 4.5	44.0%	estimated ± 3.8 pp, medium confidence
62	Kimi K2.6	43.9%	measured
63	Qwen 3.6 Max (preview)	43.9%	measured
64	Qwen3.6 Plus	43.9%	measured
65	Grok 4.3	43.8%	estimated ± 3.8 pp, medium confidence
66	GPT-5.4 nano	43.8%	estimated ± 3.8 pp, medium confidence
67	Mistral Medium 3.5 128B	43.6%	estimated ± 3.8 pp, medium confidence
68	Mercury 2.5	43.4%	estimated ± 3.8 pp, medium confidence
69	Gemini 3.1 Flash-Lite	43.4%	estimated ± 3.8 pp, medium confidence
70	Laguna M.1	43.4%	estimated ± 3.8 pp, medium confidence
71	Laguna XS.2	43.2%	estimated ± 3.8 pp, medium confidence
72	GLM-5	43.2%	measured
73	MiMo-V2.5-Pro	43.2%	measured
74	Hy3 Preview	43.1%	estimated ± 4.0 pp, medium confidence
75	Qwen3.5 397B	42.7%	estimated ± 4.0 pp, medium confidence
76	Qwen3.6-35B-A3B	42.6%	estimated ± 4.0 pp, medium confidence
77	Kimi K2.5	42.4%	estimated ± 4.0 pp, medium confidence
78	Kimi K2.5 (Reasoning)	42.4%	estimated ± 4.0 pp, medium confidence
79	Claude Sonnet 4.5	42.3%	estimated ± 4.0 pp, medium confidence
80	Qwen3.5-122B-A10B	42.2%	estimated ± 4.0 pp, medium confidence
81	MAI-Thinking-1	41.8%	estimated ± 4.0 pp, medium confidence
82	Qwen3.5-27B	41.5%	estimated ± 4.0 pp, medium confidence
83	GLM-4.7	41.5%	estimated ± 4.0 pp, medium confidence
84	Qwen3.5-35B-A3B	41.4%	estimated ± 4.0 pp, medium confidence
85	Laguna XS 2.1	41.3%	estimated ± 4.0 pp, medium confidence
86	LongCat-Flash-Lite-Sparse	41.2%	estimated ± 4.0 pp, medium confidence
87	Solar Open 2	28.3%	measured

Terminal-Bench Hard leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General