Agentic · terminal

Terminal-Bench 2.1 leaderboard

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

As of 2026-10-07, the highest measured score on Terminal-Bench 2.1 is 92.8% by SWE-2. 40 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	SWE-2	92.8%	measured
2	GPT-5.6 Sol	91.9%	measured
3	Claude Opus 5	91.8%	estimated ± 6.6 pp, low confidence
4	DeepSeek V4.1 Flash	90.6%	measured
5	MiMo-V2.6-Pro	89.9%	measured
6	Claude Sonnet 5.5	89.9%	estimated ± 3.9 pp, medium confidence
7	Claude Opus 5.5	89.8%	estimated ± 3.9 pp, medium confidence
8	GPT-6 Astra	89.7%	estimated ± 3.9 pp, medium confidence
9	Gemini 4 Argon	89.7%	estimated ± 3.9 pp, medium confidence
10	GPT-6.1 Sol	89.6%	estimated ± 3.9 pp, medium confidence
11	Claude Fable 5.1	89.5%	estimated ± 3.9 pp, medium confidence
12	Gemini 3.8 Flash	89.4%	measured
13	GPT-6 Sol	89.1%	estimated ± 3.9 pp, medium confidence
14	Muse Spark 1.3	88.8%	measured
15	Claude Haiku 5.5	88.4%	estimated ± 3.9 pp, high confidence
16	Kimi K3	88.3%	measured
17	GLM-5.3	88.2%	measured
18	Claude Mythos 5	88.0%	measured
19	DeepSeek V4 Pro 0813	87.9%	measured
20	Mistral Large 4	87.7%	estimated ± 3.9 pp, high confidence
21	MiMo-V2.6-Flash	87.6%	measured
22	Grok 4.7	87.6%	estimated ± 3.9 pp, high confidence
23	GPT-5.6 Terra	87.4%	measured
24	Qwen3.8 Max	86.6%	measured
25	Ornith-1.5-397B	86.1%	measured
26	GPT-5.5	85.8%	estimated ± 6.9 pp, medium confidence
27	Gemini 3.7 Flash	85.8%	measured
28	Grok 4.6	85.8%	estimated ± 6.6 pp, low confidence
29	Hy4 preview	85.4%	measured
30	Gemini 3.6 Flash	85.2%	estimated ± 6.9 pp, medium confidence
31	Step 5 Preview	85.0%	measured
32	GPT-5.6 Luna	84.7%	measured
33	Claude Fable 5	84.3%	measured
34	GLM-5.3-Flash	84.3%	measured
35	Gemini 3.1 Pro	84.2%	estimated ± 6.9 pp, medium confidence
36	GPT-6 Luna	84.0%	estimated ± 3.9 pp, high confidence
37	Claude Opus 4.7	83.3%	estimated ± 6.9 pp, medium confidence
38	Grok 4.5	83.3%	measured
39	Muse Spark 1.2	82.9%	measured
40	DeepSeek V4 Flash 0731	82.7%	measured
41	Kimi K2.7 Code	82.6%	estimated ± 6.9 pp, medium confidence
42	Sakana Fugu-Ultra	82.1%	measured
43	Ember-1	82.0%	measured
44	SWE-1.7	81.5%	measured
45	GLM-5.2	81.0%	measured
46	Claude Sonnet 5	80.4%	measured
47	Sakana Fugu	80.2%	measured
48	Beam	80.1%	measured
49	Muse Spark 1.1	80.0%	measured
50	Qwen3.7 Max	78.5%	estimated ± 6.9 pp, medium confidence
51	Atria Dawn Preview	78.3%	measured
52	MiMo-V2.5	78.3%	estimated ± 6.9 pp, medium confidence
53	Ornith-1.0-397B	77.5%	measured
54	Gemini 3.5 Flash	76.2%	measured
55	Composer 2.5	75.9%	estimated ± 6.9 pp, medium confidence
56	dots3-note Preview	75.1%	measured
57	Claude Sonnet 4.6	74.6%	estimated ± 6.9 pp, medium confidence
58	MiMo-V2.5-Pro	74.6%	estimated ± 6.9 pp, medium confidence
59	Claude Opus 4.8	74.6%	measured
60	GLM-5.1	74.1%	estimated ± 6.9 pp, medium confidence
61	Qwen3.8-27B	73.0%	measured
62	GPT-5.4 mini	71.0%	estimated ± 6.9 pp, medium confidence
63	Seed 2.1 Pro	71.0%	measured
64	Apodex 1.1	70.8%	measured
65	Laguna S 2.1	70.2%	measured
66	Gemini 3 Flash	69.7%	estimated ± 6.9 pp, medium confidence
67	Quasar 438B	69.3%	measured
68	Kimi K2.6	69.2%	estimated ± 6.9 pp, medium confidence
69	Qwen3.6 Plus	68.5%	estimated ± 6.9 pp, medium confidence
70	Qwen3.7 Plus	67.8%	estimated ± 6.9 pp, medium confidence
71	Ornith-1.5-35B-A3B	67.8%	measured
72	Seed 2.1 Turbo	67.6%	measured
73	MiniMax M3	66.0%	measured
74	Pokee-Isaac 28B	65.1%	measured
75	Inkling-Small	64.7%	measured
76	Ornith-1.0-35B	64.2%	measured
77	Inkling	63.8%	measured
78	MAI-Code-1.1-Flash	62.9%	measured
79	Step 3.7 Flash	59.5%	measured
80	MiniMax M2.7	59.1%	estimated ± 6.9 pp, medium confidence
81	Ling 3.0 Flash	57.0%	measured
82	Solar Pro 4	57.0%	measured
83	Nemotron 3 Ultra	56.4%	measured
84	Gemini 3.5 Flash-Lite	54.0%	measured
85	Ternary Bonsai 2 27B	52.8%	measured
86	Muse Glimmer 30B	51.7%	measured
87	Grok 4.20	46.8%	estimated ± 6.9 pp, low confidence
88	Ornith-1.5-9B	46.2%	measured
89	Claude Haiku 4.5	45.6%	estimated ± 6.9 pp, low confidence
90	K-EXAONE 2.0	43.8%	measured
91	Ornith-1.0-9B	43.1%	measured
92	Grok 4.3	39.7%	estimated ± 6.9 pp, low confidence
93	GPT-5.4 nano	38.8%	estimated ± 6.9 pp, low confidence
94	A.X K2	36.0%	measured
95	Mistral Medium 3.5 128B	30.6%	estimated ± 6.9 pp, low confidence
96	Granite 4.2 30B	29.2%	measured
97	Nemotron 3.5 Lightning 30B A3B NVFP4	23.5%	measured
98	Granite 4.2 8B	20.6%	measured
99	Mercury 2.5	17.9%	estimated ± 6.9 pp, low confidence
100	Gemini 3.1 Flash-Lite	16.9%	estimated ± 6.9 pp, low confidence
101	Laguna M.1	16.9%	estimated ± 6.9 pp, low confidence
102	MiniCPM5-2B	8.6%	measured
103	Laguna XS.2	3.8%	estimated ± 6.9 pp, low confidence
104	Command A+	0.4%	estimated ± 6.9 pp, low confidence

Terminal-Bench 2.1 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General