Terminal-Bench 2.1

Terminal-Bench 2.1 is an updated release of the Terminal-Bench benchmark that tests AI agents' ability to operate a computer via the terminal. It evaluates how well models handle real-world, end-to-end tasks autonomously, including compiling code, training models, setting up servers, system administration, data science workflows, and security tasks.

id terminal-bench-2.1 · max 1 · 42 models reported

#	Model	Score
1	DeepSeek-V4.1-Flash DeepSeek · open	0.91
2	MiMo-V2.6-Pro Xiaomi · open	0.90
3	Gemini 3.8 Flash Google	0.89
4	GPT-5.6 Sol OpenAI	0.89
5	Muse Spark 1.3 Meta	0.89
6	Kimi K3 Moonshot AI · open	0.88
7	GLM-5.3 Zhipu AI · open	0.88
8	DeepSeek-V4-Pro-0813 DeepSeek · open	0.88
9	MiMo-V2.6-Flash Xiaomi · open	0.88
10	GPT-5.6 Terra OpenAI	0.87
11	Qwen3.8 Max Alibaba Cloud / Qwen Team · open	0.87
12	Gemini 3.7 Flash Google	0.86
13	Hy4 preview Tencent · open	0.85
14	GPT-5.6 Luna OpenAI	0.85
15	Claude Fable 5 Anthropic	0.84
16	GLM-5.3-Flash Zhipu AI · open	0.84
17	DeepSeek-V4-Flash-Vision-Exp DeepSeek	0.84
18	Grok 4.5 xAI	0.83
19	Muse Spark 1.2 Meta	0.83
20	DeepSeek-V4-Flash-0731 DeepSeek · open	0.83
21	GLM-5.2 Zhipu AI · open	0.83
22	Muse Spark 1.1 Meta	0.80
23	Atria Dawn Preview Shanghai AI Laboratory · open	0.78
24	Gemini 3.6 Flash Google	0.78
25	Qwen3.8-27B Alibaba Cloud / Qwen Team · open	0.73
26	Hy3 Tencent · open	0.72
27	Seed 2.1 Pro ByteDance	0.71
28	Laguna S 2.1 Poolside · open	0.70