Which LLM scores highest on Terminal-Bench 2.1?

DeepSeek V4.1 Flash at 90.6%, then Fugu Max and Muse Spark 1.3. Scores are the pass rate as published, not a composite.

What is the difference between Terminal-Bench 2.1 and Terminal-Bench Hard?

Hard is the harder subset, and scores on it run well below the 2.1 figure for the same model. 69 models here publish a Hard number against 57 publishing 2.1, so the two columns are never averaged together.

Who reports these Terminal-Bench numbers?

Artificial Analysis measures them independently, which is why they take precedence over a provider's own figure where the two disagree. Rows measured at a reasoning effort other than the default a plain API call buys are marked as such.

The Best LLMs on Terminal-Bench 2.1

Which model solves the most Terminal-Bench 2.1 tasks?

57 of the 152 buyable models publish a Terminal-Bench 2.1 pass rate. 69 also publish Terminal-Bench Hard, shown alongside.

Method: Published Terminal-Bench 2.1 pass rate, descending, exactly as reported. Nothing is weighted, blended or rescaled here: this is one suite, reported as published, with price alongside it rather than folded into it. Pricing as of October 2026. 12 rows below carry a score measured at a non-default reasoning tier and are marked as such; every affected model is named on the hub. Read the full method.

Rank	Model	Terminal-Bench 2.1	TB Hard	AA coding	Reported by	Blended $/1M	Context
#1	DeepSeek V4.1 FlashDeepSeek	90.6	not published	not published	Vendor-reported	$0.263	1M
#2	Fugu MaxSakana AI	89.5	not published	not published	Vendor-reported	$3.00	1M
#3	Muse Spark 1.3Meta	88.8	not published	not published	Vendor-reported	$2.00	1.05M
#4	GPT-5.6 SolOpenAI	86.1	62.9	76.3	Independent	$8.00	1.05M
#5	Claude Opus 5Anthropic	86.1	not published	74.3	Unstated / mixed	$10.00	1M
#6	Kimi K3Moonshot	85.0	not published	76.2	Independent	$6.00	1.05M
#7	Claude Opus 4.8Anthropic	84.6	58.3	74.3	Unstated / mixed	$10.00	1M
#8	Claude Fable 5Anthropic	84.6	62.9	76.5	Independent	$20.00	1M
#9	Claude Opus 4.7Anthropic· non-default tier	83.1	51.5	73.6	Independent	$10.00	1M
#10	Grok 4.5xAI	81.6	not published	72.4	Independent	$3.00	500K
#11	Claude Sonnet 5Anthropic· non-default tier	80.5	not published	71.5	Unstated / mixed	$4.00	1M
#12	GPT-5.5OpenAI	80.5	57.6	71.5	Independent	$11.25	1.05M
#13	Muse Spark 1.1Meta	77.9	not published	71.3	Independent	$2.00	1M
#14	GLM-5.2Zhipu· non-default tier	77.9	50.8	68.8	Independent	$2.15	1M
#15	Gemini 3.6 FlashGoogle	77.5	not published	69.2	Independent	$1.50	1.05M
#16	Gemini 3.5 FlashGoogle	76.2	39.4	not published	Unstated / mixed	$3.38	1.05M
#17	Qwen3.7 MaxAlibaba	74.5	50.8	66.0	Independent	$3.75	1M
#18	GPT-5.6 TerraOpenAI	72.3	not published	64.7	Independent	$4.50	1.05M
#19	Claude Sonnet 4.6Anthropic· non-default tier	71.2	53.0	63.0	Independent	$6.00	1M
#20	Laguna S 2.1Poolside	70.2	not published	not published	Vendor-reported	$0.125	1.05M
#21	Nex-N2-ProNex AGI	67.8	not published	59.1	Unstated / mixed	$1.00	262K
#22	Kimi K2.7 CodeMoonshot	67.4	44.7	60.8	Independent	$1.71	262K
#23	Kimi K2.6Moonshot	65.9	43.9	61.8	Independent	$1.71	262K
#24	MiniMax M3MiniMax	65.2	42.4	58.6	Independent	$0.525	1M
#25	MiMo-V2.5-ProXiaomi	65.2	43.2	60.2	Independent	$0.544	1.05M
#26	DeepSeek V4-ProDeepSeek· non-default tier	64.8	41.7	58.7	Independent	$0.990	1M
#27	GLM-5.1Zhipu· non-default tier	61.8	43.2	55.8	Independent	$2.15	200K
#28	Qwen3.6-PlusAlibaba	61.4	43.9	54.5	Independent	$1.13	1M
#29	Qwen3.7 PlusAlibaba	61.0	47.0	55.9	Independent	$0.700	1M
#30	KAT-Coder-Pro v2.5Kwaipilot	60.7	not published	not published	Vendor-reported	$1.29	262K
#31	MiniMax M2.7MiniMax	55.4	39.4	52.6	Independent	$0.525	205K
#32	Gemini 3.5 Flash-LiteGoogle	53.6	not published	49.3	Independent	$0.850	1.05M
#33	GPT-5.6 LunaOpenAI	53.2	not published	50.7	Independent	$0.450	1.05M
#34	GPT-5.1OpenAI· non-default tier	52.4	45.5	49.4	Independent	$3.44	400K
#35	Muse Glimmer 30BMeta	51.7	not published	not published	Vendor-reported	$0.637	131K
#36	Mistral Medium 3.5Mistral	50.6	33.3	46.9	Unstated / mixed	$3.00	256K
#37	LongCat-2.0Meituan	50.2	not published	45.3	Independent	$1.30	1M
#38	Kimi K2.5Moonshot· non-default tier	45.7	34.8	46.8	Independent	$1.20	262K
#39	GLM-4.7Zhipu· non-default tier	45.3	31.8	45.3	Independent	$1.00	200K
#40	Step 3.7 FlashStepFun	39.3	35.6	39.6	Independent	$0.438	262K
#41	Qwen3 Coder NextAlibaba	38.2	18.2	36.2	Unstated / mixed	$0.290	262K
#42	Gemini 3.1 Flash-LiteGoogle	31.1	24.2	34.7	Independent	$0.563	1.05M
#43	Qwen3.5-9BAlibaba· non-default tier	29.2	24.2	28.7	Independent	$0.113	262K
#44	Gemini 2.5 ProGoogle	28.5	26.5	33.3	Independent	$3.44	1.05M
#45	Gemma 4 12BGoogle· non-default tier	27.3	18.2	31.0	Independent	$0.00	262K
#46	Mercury 2Inception Labs	27.3	26.5	31.1	Independent	$0.375	128K
#47	Command ACohere	22.8	25.0	27.8	Independent	$4.38	128K
#48	Command A+Cohere	22.8	25.0	27.8	Independent	$4.38	128K
#49	Mistral Small 4Mistral· non-default tier	21.0	17.4	26.6	Independent	$0.263	256K
#50	Mistral Large 3Mistral	12.0	15.9	20.1	Independent	$0.750	262K
#51	GPT-4.1 MiniOpenAI	10.1	7.6	20.2	Independent	$0.700	1.05M
#52	Ministral 3 14BMistral	9.7	4.5	14.4	Independent	$0.200	262K
#53	Llama 4 MaverickMeta	7.9	6.8	16.3	Independent	$0.415	1.05M
#54	GPT-4o MiniOpenAI	5.6	not published	11.4	Independent	$0.263	128K
#55	Ministral 3 8BMistral	4.1	4.5	9.7	Independent	$0.150	262K
#56	Llama 4 ScoutMeta	3.7	1.5	8.2	Independent	$0.135	1.05M
#57	GPT-4.1 NanoOpenAI	3.7	3.8	11.1	Independent	$0.175	1.05M

Frequently asked

Nearby questions, answered elsewhere

The hard subset

More models publish Hard than publish 2.1, and the two orders disagree. Scores there run well below the figures here.

SWE-bench Verified

The other pass-rate suite, published by far fewer models.

All rankings

The hub and the method The Best LLMs by Benchmark Score The Best Value LLMs per Dollar The Best LLMs by Context Window The Cheapest LLMs With a 1M-Token Context Window The Best LLMs on Terminal-Bench Hard The Best LLMs on SWE-bench Verified The Best LLMs on GPQA Diamond The Best LLMs for Tool Use on tau2-bench