Terminal-Bench 2.1 (Vals)

Terminal-Bench 2.1 in Vals AI's fixed harness.

Published by

Vals AI

Category

Agents & tools

Index weight

0.5

Models

61

Data as of

9 Sept 2026

Vals AI (vals.ai).

1. 1GPT-6 Astra87.3%
2. 2GPT-5.6 Sol85.8%
3. 3Claude Fable 5.185.0%
4. 4Claude Opus 584.6%
5. 5Gemini 3.8 Flash81.3%
6. 6Kimi K380.9%
7. 7Claude Fable 580.5%
8. 8GPT-5.6 Luna79.0%
9. 9Muse Spark 1.379.0%
10. 10Grok 4.678.3%
11. 11Gemini 3.7 Flash77.5%
12. 12GPT-5.6 Terra77.5%
13. 13GPT-5.576.4%
14. 14Claude Sonnet 574.5%
15. 15Gemini 3.5 Flash74.2%

SearchAll providersAlibabaAnthropicCursorDeepSeekGoogleInclusionAIMetaMiniMaxMistral AIMoonshot AINVIDIAOpenAIPoolsideThinking MachinesxAIXiaomiZhipu AIOpen weights61 of 61

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	GPT-6 AstraOpenAI	87.3%	–	70.6	–
2	GPT-5.6 SolOpenAI	85.8%	–	65.5	–
3	Claude Fable 5.1Anthropic	85.0%	–	70.0	–
4	Claude Opus 5Anthropic	84.6%	–	70.0	–
5	Gemini 3.8 FlashGoogle	81.3%	–	60.1	–
6	Kimi K3Moonshot AIopen	80.9%	kimi/kimi-k3	66.5	–
7	Claude Fable 5Anthropic	80.5%	–	70.4	–
8	GPT-5.6 LunaOpenAI	79.0%	–	56.5	–
9	Muse Spark 1.3Meta	79.0%	–	–	–
10	Grok 4.6xAI	78.3%	–	58.7	–
11	Gemini 3.7 FlashGoogle	77.5%	–	60.8	–
12	GPT-5.6 TerraOpenAI	77.5%	–	58.0	–
13	GPT-5.5OpenAI	76.4%	–	66.3	–
14	Claude Sonnet 5Anthropic	74.5%	–	61.2	–
15	Gemini 3.5 FlashGoogle	74.2%	–	60.4	–
16	Gemini 3.6 FlashGoogle	73.8%	–	58.6	–
17	Muse Spark 1.3Meta	72.3%	–	63.9	–
18	Claude Opus 4.8Anthropic	71.9%	–	64.7	–
19	GLM-5.3Zhipu AIopen	71.5%	–	59.9	–
20	Gemini 3.1 ProGoogle	70.8%	–	60.7	–
21	Muse Spark 1.2Meta	69.7%	–	61.3	–
22	Claude Opus 4.8 Claude CodeAnthropic	69.7%	–	–	–
23	Muse Spark 1.1Meta	69.3%	–	56.4	–
24	Claude Opus 4.7Anthropic	68.5%	–	65.5	–
25	GLM-5.2Zhipu AIopen	67.8%	–	57.6	–
26	Grok 4.5xAI	67.8%	–	54.2	–
27	Qwen3 8Alibaba	67.4%	–	63.9	–
28	DeepSeek V4 FlashDeepSeekopen	67.0%	0731	59.1	–
29	Kimi K2.7 CodeMoonshot AIopen	67.0%	–	56.2	–
30	GLM-5.3-FlashZhipu AIopen	62.9%	–	55.6	–
31	Qwen3 7Alibaba	61.0%	–	58.9	–
32	Qwen 3.8 27BAlibabaopen	58.4%	–	52.0	–
33	CComposer 2.5Cursor	58.4%	cursor/composer-2.5	–	–
34	Claude Sonnet 4.6Anthropic	57.3%	–	60.3	–
35	XMiMo-V2.5-ProXiaomiopen	57.3%	–	59.0	–
36	GPT 5.5 CodexOpenAI	57.3%	–	–	–
37	GPT 5.5 FactoryOpenAI	57.3%	–	–	–
38	GLM-5.1Zhipu AIopen	56.9%	zai/glm-5.1	60.1	–
39	TInkling-SmallThinking Machinesopen	55.1%	–	55.0	–
40	DeepSeek V4 ProDeepSeekopen	54.7%	max, 0813	56.1	–
41	GPT-5.4 miniOpenAI	54.7%	–	52.9	–
42	Gemini 3 FlashGoogle	53.9%	–	58.5	–
43	Kimi K2.6Moonshot AIopen	53.6%	kimi/kimi-k2.6	60.6	–
44	MiniMax-M3MiniMaxopen	53.6%	minimax/MiniMax-M3	56.9	–
45	Qwen3.6 PlusAlibabaopen	53.2%	–	59.0	–
46	Qwen3.7 PlusAlibaba	52.8%	–	59.7	–
47	Nemotron 3 Ultra 550B A55BNVIDIAopen	50.9%	–	58.3	–
48	DeepSeek V4 ProDeepSeekopen	50.2%	–	59.7	–
49	ILing 3.0 FlashInclusionAIopen	50.2%	2607	54.0	–
50	Gemini 3.5 Flash LiteGoogle	50.2%	–	44.0	–
51	MiniMax-M2.7MiniMaxopen	48.7%	minimax/MiniMax-M2.7	56.5	–
52	Grok 4.20xAI	44.2%	–	54.3	–
53	Claude Haiku 4.5Anthropic	43.8%	–	48.0	–
54	Kimi K2.5Moonshot AIopen	42.0%	–	55.3	–
55	Grok 4.3xAI	42.0%	–	50.1	–
56	GPT-5.4 nanoOpenAI	41.6%	–	48.3	–
57	Mistral Medium 3.5Mistral AI	39.0%	–	–	–
58	Gemini 3.1 Flash LiteGoogle	34.1%	–	49.1	–
59	Laguna M.1Poolsideopen	34.1%	poolside/laguna-m.1	37.5	–
60	Laguna XS.2Poolside	25.8%	poolside/laguna-xs.2	36.3	–

Show 1 more