Agentic · terminal

Terminal-Bench 2.0 leaderboard

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

As of 2026-10-07, the highest measured score on Terminal-Bench 2.0 is 82.0% by GPT-5.5. 43 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	84.0%	estimated ± 5.9 pp, low confidence
2	GPT-5.6 Sol	83.2%	estimated ± 5.9 pp, low confidence
3	Claude Fable 5.1	82.7%	estimated ± 5.9 pp, low confidence
4	Claude Opus 5	82.5%	estimated ± 5.9 pp, low confidence
5	GPT-5.5	82.0%	measured
6	Gemini 3.8 Flash	80.6%	estimated ± 5.9 pp, low confidence
7	Kimi K3	80.4%	estimated ± 5.9 pp, low confidence
8	Claude Fable 5	80.1%	estimated ± 5.9 pp, low confidence
9	GPT-5.6 Luna	79.2%	estimated ± 5.9 pp, low confidence
10	Grok 4.6	78.8%	estimated ± 5.9 pp, low confidence
11	Gemini 3.7 Flash	78.3%	estimated ± 5.9 pp, low confidence
12	GPT-5.6 Terra	78.3%	estimated ± 5.9 pp, low confidence
13	GPT-5.3 Codex	77.3%	measured
14	Claude Sonnet 5	76.4%	estimated ± 5.9 pp, medium confidence
15	DeepSeek V4.1 Flash	76.4%	estimated ± 5.9 pp, medium confidence
16	Gemini 3.5 Flash	76.2%	estimated ± 5.9 pp, medium confidence
17	Gemini 3.6 Flash	76.0%	estimated ± 5.9 pp, medium confidence
18	Grok 4.7	75.7%	estimated ± 5.9 pp, medium confidence
19	GPT-5.4	75.1%	measured
20	Muse Spark 1.3	75.0%	estimated ± 5.9 pp, medium confidence
21	Claude Opus 4.8	74.8%	estimated ± 5.9 pp, medium confidence
22	GLM-5.3	74.5%	estimated ± 5.9 pp, medium confidence
23	Gemini 3.1 Pro	74.1%	estimated ± 5.9 pp, medium confidence
24	Muse Spark 1.2	73.3%	estimated ± 5.9 pp, medium confidence
25	Muse Spark 1.1	73.1%	estimated ± 5.9 pp, medium confidence
26	Claude Opus 4.7	72.5%	estimated ± 5.9 pp, medium confidence
27	GLM-5.2	72.1%	estimated ± 5.9 pp, medium confidence
28	Grok 4.5	72.1%	estimated ± 5.9 pp, medium confidence
29	Qwen3.8 Max	71.8%	estimated ± 5.9 pp, medium confidence
30	Kimi K2.7 Code	71.5%	estimated ± 5.9 pp, medium confidence
31	Qwen3.7 Plus	70.3%	measured
32	Qwen3.7 Max	69.7%	measured
33	Claude Opus 4.7 (Adaptive)	69.4%	measured
34	Composer 2.5	69.3%	measured
35	GLM-5.3-Flash	68.6%	estimated ± 5.9 pp, medium confidence
36	MiMo-V2.5-Pro	68.4%	measured
37	DeepSeek V4 Pro 0813	67.9%	measured
38	Kimi K2.6	66.7%	measured
39	MiMo-V2.5	65.8%	measured
40	Claude Opus 4.6	65.4%	measured
41	Qwen 3.6 Max (preview)	65.4%	measured
42	Qwen3.8-27B	65.3%	estimated ± 5.9 pp, medium confidence
43	GLM-5.1	63.5%	measured
44	Hy4 preview	62.8%	estimated ± 5.9 pp, medium confidence
45	Inkling-Small	62.8%	estimated ± 5.9 pp, medium confidence
46	Gemini 3 Flash	61.9%	estimated ± 5.9 pp, medium confidence
47	Composer 2	61.7%	measured
48	MiniMax M3	61.6%	estimated ± 5.9 pp, medium confidence
49	Qwen3.6 Plus	61.6%	measured
50	GPT-5.4 mini	60.0%	measured
51	Nemotron 3 Ultra	59.5%	estimated ± 5.9 pp, medium confidence
52	Claude Opus 4.5	59.3%	measured
53	Qwen3.6-27B	59.3%	measured
54	Claude Sonnet 4.6	59.1%	measured
55	Muse Spark	59.0%	measured
56	Gemini 3.5 Flash-Lite	58.9%	estimated ± 5.9 pp, medium confidence
57	Ling 3.0 Flash	58.9%	estimated ± 5.9 pp, medium confidence
58	MiniMax M2.7	57.0%	measured
59	DeepSeek V4 Flash 0731	56.9%	measured
60	Inkling	56.7%	estimated ± 5.9 pp, medium confidence
61	GLM-5	56.2%	measured
62	Hy3 Preview	54.4%	measured
63	Claude Haiku 4.5	53.4%	estimated ± 5.9 pp, medium confidence
64	Qwen3.5 397B	52.5%	measured
65	Grok 4.3	51.7%	estimated ± 5.9 pp, medium confidence
66	Qwen3.6-35B-A3B	51.5%	measured
67	Kimi K2.5	50.8%	measured
68	Kimi K2.5 (Reasoning)	50.8%	measured
69	Claude Sonnet 4.5	50.0%	measured
70	Qwen3.5-122B-A10B	49.4%	measured
71	Mistral Medium 3.5 128B	49.0%	estimated ± 5.9 pp, medium confidence
72	Grok 4.20	47.1%	measured
73	GPT-5.4 nano	46.3%	measured
74	MAI-Thinking-1	46.0%	measured
75	Laguna M.1	45.8%	measured
76	Mercury 2.5	44.6%	estimated ± 5.9 pp, medium confidence
77	Gemini 3.1 Flash-Lite	44.2%	estimated ± 5.9 pp, medium confidence
78	Qwen3.5-27B	41.6%	measured
79	GLM-4.7	41.0%	measured
80	Qwen3.5-35B-A3B	40.5%	measured
81	Laguna XS 2.1	37.5%	measured
82	Laguna XS.2	35.7%	measured
83	LongCat-Flash-Lite-Sparse	33.7%	measured
84	Command A+	25.5%	estimated ± 5.9 pp, low confidence

Terminal-Bench 2.0 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General