Terminal-Bench 4.0

A harder 66-task benchmark of complex terminal work across software, machine learning, science, operations, security, hardware, and media, with recalibrated compute and time allowances and improved instructions, environments, and verifiers.

Category: Coding
Intelligence Index weight: 10%
Models: 147 current, 252 in total
Median: 11.1%
Range: 0.0%–67.2%

Score distribution

Lowest: 0.0%
Median: 11.1%
Highest: 67.2%

Leaderboard

Include deprecated models

Rank	Model	Score	Cost / task	Time / task	Output tokens / task
1	Kimi K3.1 Pro Kimi	67.2%	$5.27	59m 27s	116K
2	Gemini 4 Carbon (High) Google	65.2%	$10.05	32m 35s	161K
3	Claude Sonnet 5.5 (Max, Default Fallback) Anthropic	63.6%	$12.65	41m 30s	382K
4	Gemini 4 Carbon (Low) Google	62.6%	$8.57	35m 49s	139K
5	Claude Fable 5.2 (Xhigh, Default Fallback) Anthropic	61.6%	$20.81	41m 57s	260K
6	Gemini 4 Carbon (Medium) Google	60.6%	$8.82	32m 18s	143K
7	Claude Fable 5.2 (Max, Default Fallback) Anthropic	59.6%	$31.59	61m 7s	377K
8	Claude Fable 5.2 (High, Default Fallback) Anthropic	59.6%	$13.47	28m 13s	158K
9	Claude Opus 5.5 (Max, Default Fallback) Anthropic	59.6%	$13.11	36m 4s	253K
10	Claude Opus 5.5 (Xhigh, Default Fallback) Anthropic	59.6%	$8.78	26m 3s	169K
11	GPT-6 Astra (Xhigh) OpenAI	59.6%	$5.86	18m 32s	43.3K
12	GPT-6 Astra (Max) OpenAI	59.1%	$8.50	33m 31s	75.7K
13	Claude Sonnet 5.5 (Xhigh, Default Fallback) Anthropic	57.1%	$6.08	23m 2s	192K
14	Gemini 4 Argon (High) Google	57.1%	$13.56	40m 39s	165K
15	Muse Wonder Full (Max) Meta	57.1%	—	—	153K
16	Claude Opus 5.5 (High, Default Fallback) Anthropic	56.6%	$5.12	15m 56s	100K
17	GPT-6.1 Sol (Max) OpenAI	56.1%	$1.82	32m 5s	108K
18	GPT-6.1 Terra (Max) OpenAI	55.6%	$1.49	23m 2s	101K
19	Claude Fable 5.2 (Medium, Default Fallback) Anthropic	55.1%	$5.25	11m 36s	63.2K
20	Claude Fable 5.2 (Low, Default Fallback) Anthropic	54.5%	$4.34	9m 30s	52.9K
21	GPT-6.1 Sol (Xhigh) OpenAI	54.0%	$1.03	17m 0s	56.9K
22	GPT-6 Astra (High) OpenAI	54.0%	$4.05	12m 54s	30.4K
23	Claude Opus 5.5 (Medium, Default Fallback) Anthropic	52.5%	$4.04	12m 12s	75.7K
24	Muse Wonder (Max) Meta	52.0%	$12.21	22m 52s	167K
25	GPT-6.1 Sol (High) OpenAI	51.5%	$0.83	12m 46s	43.7K
26	GPT-6 Astra (Medium) OpenAI	49.5%	$4.43	11m 35s	29.7K
27	GPT-6.1 Sol (Medium) OpenAI	48.0%	$0.61	9m 2s	32K
28	GPT-6.1 Terra (Xhigh) OpenAI	46.5%	$0.66	11m 1s	47.5K
29	Claude Sonnet 5.5 (High, Default Fallback) Anthropic	43.9%	$2.19	9m 4s	76.7K
30	GLM-5.3 (Max) Z AI	41.9%	$8.06	47m 13s	195K
31	GPT-6.1 Terra (High) OpenAI	41.9%	$0.44	7m 14s	30.4K
32	GPT-6 Astra (Low) OpenAI	41.9%	$2.25	6m 15s	14.9K
33	Muse Ember (Max) Meta	40.9%	$0.27	5m 25s	148K
34	Qwen3.8 Max (0902) Alibaba	38.9%	$18.68	95m 29s	293K
35	Muse Wonder (High) Meta	38.4%	$8.10	15m 55s	116K
36	Tencent Hy4 Tencent	37.4%	$1.64	—	123K
37	Muse Ember (Xhigh) Meta	35.4%	$0.22	4m 48s	132K
38	GLM-5.3 (Low) Z AI	34.8%	$6.40	36m 7s	128K
39	MiMo-V2.6-Pro Xiaomi	34.8%	$0.44	54m 30s	184K
40	DeepSeek V4.1 Flash 1017 (Max) DeepSeek	33.8%	$0.88	26m 20s	218K
41	Ling 3.1 Flash InclusionAI	33.3%	—	—	—
42	Step 5 Preview StepFun	33.3%	$5.18	30m 6s	260K
43	Claude Haiku 5.5 (Max) Anthropic	32.8%	$0.65	22m 44s	347K
44	GLM-5.3-Flash Z AI	32.8%	$0.89	60m 57s	181K
45	Ling 3.1 Flash 1019 InclusionAI	32.8%	$6.59	—	434K
46	GPT-6.1 Terra (Medium) OpenAI	32.3%	$0.28	4m 38s	19.3K
47	Claude Opus 5.5 (Low, Default Fallback) Anthropic	31.3%	$2.08	5m 49s	37.8K
48	GPT-6.1 Sol (Low) OpenAI	30.8%	$0.38	5m 7s	17.4K
49	Claude Sonnet 5.5 (Medium, Default Fallback) Anthropic	29.8%	$1.58	6m 11s	51.6K
50	Nanbeige4.3-33B Nanbeige	29.8%	—	—	313K

1
Kimi K3.1 Pro
Kimi
67.2%
$5.27
59m 27s
116K
2
Gemini 4 Carbon (High)

Data from Artificial Analysis

Command Palette

Terminal-Bench 4.0

Score distribution

Leaderboard