Benchmarks/Terminal Bench

Terminal Bench

Terminal Bench — tests the ability to accomplish real-world tasks using terminal commands, evaluating shell scripting and CLI tool proficiency.

coding

Models Tested

78.4

Top Score

40.3

Average Score

Rankings

#	Model	Score
1	Gemini 3.1 Pro PreviewGoogle DeepMind	78.4
2	Claude Opus 4.6Anthropic	69.9
3	GPT-5.2 ChatOpenAI	64.9
4	GPT-5.2OpenAI	64.9
5	Gemini 3 Flash PreviewGoogle DeepMind	64.3
6	Claude Opus 4.5Anthropic	63.1
7	Gemini 3 ProGoogle	61.8
8	GPT-5 ChatOpenAI	49.6
9	GPT-5OpenAI	49.6
10	GPT-5.1OpenAI	47.6
11	GPT-5.1 ChatOpenAI	47.6
12	Claude Sonnet 4.5Anthropic	42.8
13	Claude Opus 4.1Anthropic	38.0
14	M Kimi K2 Thinkingmoonshotai	35.7
15	Gemini 2.5 ProGoogle DeepMind	32.6
16	GPT-5 MiniOpenAI	31.9
17	Claude Haiku 4.5Anthropic	29.8
18	M Kimi K2 0711moonshotai	27.8
19	Grok 4xAI	27.2
20	Qwen3 Coder 480B A35B (free)Alibaba Qwen	25.4
21	Qwen3 Coder 480B A35BAlibaba Qwen	25.4
22	ZA GLM 4.6z-ai	24.5
23	gpt-oss-120b (free)OpenAI	18.7
24	gpt-oss-120bOpenAI	18.7
25	gpt-oss-20b (free)OpenAI	18.7
26	gpt-oss-20bOpenAI	18.7
27	GPT-5 NanoOpenAI	11.5