Terminal-Bench Hard

Hard subset of Terminal-Bench in a fixed harness. Run by Artificial Analysis.

Published by

Artificial Analysis

Category

Agents & tools

Index weight

0.5

Models

387

Data as of

9 Sept 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

1. 1GPT-5.6 Sol65.9%
2. 2Claude Fable 562.9%
3. 3GPT-5.6 Sol62.9%
4. 4GPT-5.6 Terra62.9%
5. 5GPT-5.6 Sol62.1%
6. 6GPT-5.6 Sol61.4%
7. 7GPT-5.560.6%
8. 8GPT-5.6 Sol60.6%
9. 9GPT-5.559.9%
10. 10Claude Opus 4.858.3%
11. 11GPT-5.557.6%
12. 12GPT-5.457.6%
13. 13GPT-5.6 Terra57.6%
14. 14GPT-5.6 Terra57.6%
15. 15Claude Opus 4.754.5%

SearchAll providersAI21 LabsAlibabaAllen Institute for AIAllenaiAmazonAnt GroupAnt-groupAnthropicArcee AIBaiduByteDanceChina MobileCohereDeep CogitoDeepSeekGoogleIBMInceptionKimiKorea TelecomKwaiKATLG AI ResearchLiquid AILongCatMBZUAI Institute of Foundation ModelsMetaMicrosoftMiniMaxMistral AIMoonshot AIMotif TechnologiesMultiverse ComputingNanbeigeNaverNous ResearchNVIDIAOpenAIOpenBMBPrime IntellectQwenReka AISarvamServiceNowStepFunSwiss AI InitiativeTII UAETrillion LabsUnknownUpstagexAIXiaomiZhipu AIOpen weights387 of 387

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	GPT-5.6 SolOpenAI	65.9%	GPT-5.6 Sol (max)	65.3	–
2	Claude Fable 5Anthropic	62.9%	Claude Fable 5 (with fallback)	70.4	–
3	GPT-5.6 SolOpenAI	62.9%	–	66.0	–
4	GPT-5.6 TerraOpenAI	62.9%	–	64.2	–
5	GPT-5.6 SolOpenAI	62.1%	–	68.0	–
6	GPT-5.6 SolOpenAI	61.4%	–	67.5	–
7	GPT-5.5OpenAI	60.6%	GPT-5.5 (xhigh)	66.6	–
8	GPT-5.6 SolOpenAI	60.6%	–	64.6	–
9	GPT-5.5OpenAI	59.9%	–	66.3	–
10	Claude Opus 4.8Anthropic	58.3%	Claude Opus 4.8 (max)	64.7	–
11	GPT-5.5OpenAI	57.6%	–	64.9	–
12	GPT-5.4OpenAI	57.6%	GPT-5.4 (xhigh)	63.4	–
13	GPT-5.6 TerraOpenAI	57.6%	–	63.1	–
14	GPT-5.6 TerraOpenAI	57.6%	GPT-5.6 Terra (max)	62.3	–
15	Claude Opus 4.7Anthropic	54.5%	–	60.1	–
16	Gemini 3.1 ProGoogle	53.8%	–	63.9	–
17	GPT-5.3 CodexOpenAI	53.0%	GPT-5.3 Codex (xhigh)	62.4	–
18	Claude Sonnet 4.6Anthropic	53.0%	–	60.3	–
19	GPT-5.5OpenAI	52.3%	–	62.3	–
20	GPT-5.4 miniOpenAI	52.3%	GPT-5.4 mini (xhigh)	54.8	–
21	Claude Opus 4.7Anthropic	51.5%	Claude Opus 4.7 (max)	65.5	–
22	Qwen3 7Alibaba	50.8%	–	58.9	–
23	GLM-5.2Zhipu AIopen	50.8%	GLM-5.2 (max)	57.4	–
24	KKAT-Coder-Pro V2KwaiKAT	49.2%	–	58.1	–
25	GPT-5.5OpenAI	49.2%	–	56.2	–
26	Claude Opus 4.6Anthropic	48.5%	–	63.1	–
27	GPT-5.2OpenAI	47.0%	GPT-5.2 (xhigh)	61.5	–
28	Claude Opus 4.5Anthropic	47.0%	–	60.6	–
29	Qwen3.7 PlusAlibaba	47.0%	Qwen3.7 Plus	59.7	–
30	DeepSeek V4 ProDeepSeekopen	46.2%	0424	59.7	–
31	Gemini 3.5 FlashGoogle	46.2%	–	57.3	–
32	Muse SparkMeta	45.5%	–	65.4	–
33	GPT-5.1OpenAI	45.5%	GPT-5.1 (high)	59.4	–
34	Kimi K2.7 CodeMoonshot AIopen	44.7%	Kimi K2.7 Code	56.2	–
35	Qwen3 6Alibaba	43.9%	–	62.0	–
36	Kimi K2.6Moonshot AIopen	43.9%	Kimi K2.6	60.6	–
37	Qwen3.6 PlusAlibabaopen	43.9%	Qwen3.6 Plus	59.0	–
38	GPT-5.6 TerraOpenAI	43.9%	–	58.8	–
39	GPT-5.4OpenAI	43.2%	–	60.4	–
40	GLM-5.1Zhipu AIopen	43.2%	GLM-5.1	60.1	–
41	GLM-5Zhipu AIopen	43.2%	GLM-5	59.0	–
42	XMiMo-V2.5-ProXiaomiopen	43.2%	–	59.0	–
43	GPT-5.2OpenAI	43.2%	–	58.5	–
44	GPT-5.5 InstantOpenAI	42.4%	–	60.4	–
45	MiniMax-M3MiniMaxopen	42.4%	MiniMax-M3	56.9	–
46	Claude Sonnet 4.6Anthropic	42.4%	–	55.6	–
47	GPT-5.4 nanoOpenAI	42.4%	GPT-5.4 nano (xhigh)	53.9	–
48	Gemini 3 ProGoogle	41.7%	Gemini 3 Pro Preview (high)	61.6	–
49	DeepSeek V4 ProDeepSeekopen	41.7%	high, 0424	60.0	–
50	XMiMo-V2.5Xiaomiopen	41.7%	0424	58.3	–
51	Gemini 3.5 FlashGoogle	40.9%	Gemini 3.5 Flash	60.7	–
52	XMimo v2 ProXiaomi	40.9%	–	60.5	–
53	Qwen3.5 397B-A17BAlibabaopen	40.9%	Qwen3.5 397B A17B	58.6	–
54	Claude Opus 4.5Anthropic	40.9%	–	58.0	–
55	Grok 4.20xAI	40.9%	0309	57.9	–
56	Gemini 3.5 FlashGoogle	39.4%	–	63.9	–
57	MiniMax-M2.7MiniMaxopen	39.4%	MiniMax-M2.7	56.5	–
58	GLM-5Zhipu AIopen	39.4%	–	54.5	–
59	Gemini 3 FlashGoogle	38.6%	–	61.5	–
60	DeepSeek V4 FlashDeepSeekopen	38.6%	high, 0420	59.3	–

Show 100 more