Terminal-Bench 2.0 leaderboard

A benchmark for agentic software engineering tasks executed in real terminal environments. DeepSeek reports it in the agentic section, while BenchLM also mirrors it in coding for models that publish it as a developer-task signal.

Models ranked by their published Terminal-Bench 2.0 value

Rank	Model	Provider	Interactive CLI agent evaluation
1	GPT-5.6 Sol	OpenAI	91.9
2	Claude Mythos 5	Anthropic	88.0
3	GPT-5.6 Terra	OpenAI	87.4
4	GPT-5.6 Luna	OpenAI	84.7
5	Claude Fable 5	Anthropic	84.3
6	Grok 4.5	xAI	83.3
7	SASakana Fugu-Ultra	Sakana AI	82.1
8	GPT-5.5	OpenAI	82.0
9	SWE-1.7	Cognition	81.5
10	GLM-5.2	Z.AI	81.0
11	Claude Sonnet 5	Anthropic	80.4
12	SASakana Fugu	Sakana AI	80.2
13	Muse Spark 1.1	Meta	80.0
14	DAOrnith-1.0-397B	DeepReinforce AI	77.5
15	Gemini 3.5 Flash	Google	76.2
16	Claude Opus 4.8	Anthropic	74.6
17	Qwen3.7 Plus	Alibaba	70.3
18	Laguna S 2.1	Poolside	70.2
19	Qwen3.7 Max	Alibaba	69.7
20	Claude Opus 4.7 (Adaptive)	Anthropic	69.4
21	Composer 2.5	Cursor	69.3
22	MiMo-V2.5-Pro	Xiaomi	68.4
23	DeepSeek V4 Pro 0813	DeepSeek	67.9
24	Kimi K2.6	Moonshot AI	66.7
25	MiniMax M3	MiniMax	66.0
26	MiMo-V2.5	Xiaomi	65.8
27	Qwen 3.6 Max (preview)	Alibaba	65.4
28	TMInkling-Small	Thinking Machines Lab	64.7
29	DAOrnith-1.0-35B	DeepReinforce AI	64.2
30	TMInkling	Thinking Machines Lab	63.8
31	Composer 2	Cursor	61.7
32	Step 3.7 Flash	StepFun	59.5
33	Qwen3.6-27B	Alibaba	59.3
34	DeepSeek V4 Flash 0731	DeepSeek	56.9
35	Nemotron 3 Ultra	NVIDIA	56.4
36	Hy3 Preview	Tencent	54.4
37	Gemini 3.5 Flash-Lite	Google	54.0
38	Qwen3.6-35B-A3B	Alibaba	51.5
39	MAI-Thinking-1	Microsoft	46.0
40	Laguna M.1	Poolside	45.8
41	DAOrnith-1.0-9B	DeepReinforce AI	43.1
42	Laguna XS 2.1	Poolside	37.5
43	Laguna XS.2	Poolside	35.7
44	LongCat-Flash-Lite-Sparse	Meituan	33.7
45	Nemotron 3.5 Lightning 30B A3B NVFP4	NVIDIA	23.5

Evidence key: Observed

Rows are ordered by the value Terminal-Bench 2.0 published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.