Agents · Benchmark profile

Terminal-Bench Hard

Harder agentic terminal evaluation suite spanning software engineering, systems administration, and data processing tasks.

Ranking weightedCurrent / rollingActive91 models

Data verified 15 July 2026 · Methodology 1.2.0

Benchmark score on Terminal-Bench Hard

Leaderboard · 91 models

Kimi K3 leads at 85%, followed by GPT-5.6 Sol (65.9%) and Claude Fable 5 (62.9%).

Ranking · 1.4%AgentsCurrent

#1Closed

Kimi K3

Moonshot AI

kimi-k3

85%

Overall #3 · Context 1M tokens

#2Closed

GPT-5.6 Sol

OpenAI

gpt-5.6-sol

65.9%

Overall #4 · Context 1.05M tokens

#3Closed

Claude Fable 5

Anthropic

claude-fable-5

62.9%

Overall #2 · Context 1M tokens

Visual analysis

See the field, not just the table.

Three views of the same sourced leaderboard: model placement, the score curve, and descriptive provider averages. These visuals never alter ranking calculations.

ModelsColours

Moonshot AIOpenAIAnthropicGoogleZ.AIAlibaba CloudDeepSeek

Leaderboard bars

20 ranked models · higher is better

View accessible chart data

Accessible data for Leaderboard bars
Model	Rank	Provider	Score
Kimi K3	#1	Moonshot AI	85%
GPT-5.6 Sol	#2	OpenAI	65.9%
Claude Fable 5	#3	Anthropic	62.9%
GPT-5.5	#4	OpenAI	60.6%
Claude Opus 4.8	#5	Anthropic	58.3%
GPT-5.4	#6	OpenAI	57.6%
GPT-5.6 Terra	#7	OpenAI	57.6%
Gemini 3.1 Pro Preview	#8	Google	53.8%
Claude Sonnet 4.6	#9	Anthropic	53.0%
GPT-5.3-Codex	#10	OpenAI	53.0%
GPT-5.4 mini	#11	OpenAI	52.3%
Claude Opus 4.7	#12	Anthropic	51.5%
GLM-5.2	#13	Z.AI	50.8%
Qwen3.7-Max	#14	Alibaba Cloud	50.8%
Claude Opus 4.5	#15	Anthropic	47.0%
GPT-5.2	#16	OpenAI	47.0%
Qwen3.7-Plus	#17	Alibaba Cloud	47.0%
Claude Opus 4.6	#18	Anthropic	46.2%
DeepSeek V4 Pro	#19	DeepSeek	46.2%
GPT-5.1	#20	OpenAI	45.5%

Score curve

Shows how quickly scores separate down the field

View accessible chart data

Accessible data for Score curve
Model	Rank	Provider	Score
Kimi K3	#1	Moonshot AI	85%
GPT-5.6 Sol	#2	OpenAI	65.9%
Claude Fable 5	#3	Anthropic	62.9%
GPT-5.5	#4	OpenAI	60.6%
Claude Opus 4.8	#5	Anthropic	58.3%
GPT-5.4	#6	OpenAI	57.6%
GPT-5.6 Terra	#7	OpenAI	57.6%
Gemini 3.1 Pro Preview	#8	Google	53.8%
Claude Sonnet 4.6	#9	Anthropic	53.0%
GPT-5.3-Codex	#10	OpenAI	53.0%
GPT-5.4 mini	#11	OpenAI	52.3%
Claude Opus 4.7	#12	Anthropic	51.5%
GLM-5.2	#13	Z.AI	50.8%
Qwen3.7-Max	#14	Alibaba Cloud	50.8%
Claude Opus 4.5	#15	Anthropic	47.0%
GPT-5.2	#16	OpenAI	47.0%
Qwen3.7-Plus	#17	Alibaba Cloud	47.0%
Claude Opus 4.6	#18	Anthropic	46.2%
DeepSeek V4 Pro	#19	DeepSeek	46.2%
GPT-5.1	#20	OpenAI	45.5%

Provider profile

Average within the visible cohort · descriptive only

View accessible chart data

Accessible data for Provider profile
Provider	Models	Average score
Moonshot AI	1	85%
OpenAI	8	54.9%
Google	1	53.8%
Anthropic	6	53.2%
Z.AI	1	50.8%
Alibaba Cloud	2	48.9%
DeepSeek	1	46.2%

Benchmark score table (91 models)

One best score per model · higher is better

Terminal-Bench Hard leaderboard ranked by score
Rank	Model	Provider	License	Score
#1	Kimi K3 kimi-k3	Moonshot AI	closed	85%
#2	GPT-5.6 Sol gpt-5.6-sol	OpenAI	closed	65.9%
#3	Claude Fable 5 claude-fable-5	Anthropic	closed	62.9%
#4	GPT-5.5 gpt-5.5	OpenAI	closed	60.6%
#5	Claude Opus 4.8 claude-opus-4-8	Anthropic	closed	58.3%
#6	GPT-5.4 gpt-5.4	OpenAI	closed	57.6%
#7	GPT-5.6 Terra gpt-5.6-terra	OpenAI	closed	57.6%
#8	Gemini 3.1 Pro Preview gemini-3.1-pro-preview	Google	closed	53.8%
#9	Claude Sonnet 4.6 claude-sonnet-4-6	Anthropic	closed	53.0%
#10	GPT-5.3-Codex gpt-5.3-codex	OpenAI	closed	53.0%
#11	GPT-5.4 mini gpt-5.4-mini	OpenAI	closed	52.3%
#12	Claude Opus 4.7 claude-opus-4-7	Anthropic	closed	51.5%
#13	GLM-5.2 glm-5-2	Z.AI	open	50.8%
#14	Qwen3.7-Max qwen3.7-max	Alibaba Cloud	—	50.8%
#15	Claude Opus 4.5 claude-opus-4-5	Anthropic	closed	47.0%
#16	GPT-5.2 gpt-5.2	OpenAI	closed	47.0%
#17	Qwen3.7-Plus qwen3.7-plus	Alibaba Cloud	—	47.0%
#18	Claude Opus 4.6 claude-opus-4-6	Anthropic	closed	46.2%
#19	DeepSeek V4 Pro deepseek-v4-pro	DeepSeek	open	46.2%
#20	GPT-5.1 gpt.5.1	OpenAI	closed	45.5%
#21	Muse Spark muse-spark	Meta	closed	45.5%
#22	Kimi K2.7 Code kimi-k2.7-code	Moonshot AI	open	44.7%
#23	Kimi K2.6 kimi-k2.6	Moonshot AI	open	43.9%
#24	Qwen3.6 Max qwen3.6-max	Alibaba Cloud	closed	43.9%
#25	Qwen3.6 Plus qwen3-6-plus	Alibaba Cloud	closed	43.9%
#26	GLM-5 glm-5	Z.AI	open	43.2%
#27	GLM-5.1 glm-5-1	Z.AI	open	43.2%
#28	MiMo-V2.5-Pro mimo-v2-5-pro	Xiaomi	closed	43.2%
#29	GPT-5.4 nano gpt.5.4.nano	OpenAI	closed	42.4%
#30	MiniMax M3 MiniMax-M3	MiniMax	—	42.4%
#31	Gemini 3 Pro gemini-3-pro	Google	closed	41.7%
#32	MiMo-V2.5 mimo-v2.5	Xiaomi	open	41.7%
#33	Gemini 3.5 Flash gemini-3.5-flash	Google	closed	40.9%
#34	MiMo-V2-Pro mimo-v2-pro	Xiaomi	closed	40.9%
#35	Qwen3.5 397B qwen3.5-397b-a17b	Alibaba Cloud	open	40.9%
#36	MiniMax M2.7 minimax-m2-7	MiniMax	open	39.4%
#37	Gemini 3 Flash gemini-3-flash	Google	closed	38.6%
#38	GPT-5 Codex gpt-5-codex	OpenAI	closed	37.9%
#39	Grok 4 grok-4	xAI	closed	37.9%
#40	Grok 4.20 grok-4.20	xAI	closed	37.9%
#41	Grok 4.3 grok-4.3	xAI	closed	37.9%
#42	GPT-5.2 Codex gpt-5.2-codex	OpenAI	closed	37.1%
#43	o3 o3	OpenAI	closed	37.1%
#44	Gemma 4 31B gemma-4-31b	Google	open	36.4%
#45	Nemotron 3 Ultra nemotron-3-ultra-550b	NVIDIA	open	36.4%
#46	Claude Sonnet 4.5 claude-sonnet-4-5	Anthropic	closed	35.6%
#47	DeepSeek V3.2 deepseek-v3.2	DeepSeek	open	35.6%
#48	DeepSeek V4 Flash deepseek-v4-flash	DeepSeek	open	35.6%
#49	Kimi K2.5 kimi-k2.5	Moonshot AI	—	34.8%
#50	MiMo-V2-Omni mimo-v2-omni	Xiaomi	closed	34.8%
#51	MiniMax M2.5 minimax-m2.5	MiniMax	open	34.8%
#52	Qwen3.6 27B qwen3.6-27b	Alibaba Cloud	open	34.8%
#53	Claude Opus 4.1 claude-opus-4-1	Anthropic	closed	34.3%
#54	GLM-5-Turbo glm-5-turbo	Z.AI	closed	33.3%
#55	Mistral Medium 3.5 mistral-medium-3-5	Mistral AI	open	33.3%
#56	GLM-5V-Turbo glm-5v-turbo	Z.AI	closed	32.6%
#57	GPT-5 gpt-5	OpenAI	closed	32.6%
#58	Qwen3.5 27B qwen3.5-27b	Alibaba Cloud	open	32.6%
#59	GLM-4.7 glm-4.7	Z.AI	open	31.8%
#60	Claude Opus 4 claude-opus-4	Anthropic	closed	31.1%
#61	Claude Sonnet 4 claude-sonnet-4	Anthropic	closed	31.1%
#62	Kimi K2 Thinking kimi-k2-thinking	Moonshot AI	open	31.1%
#63	Ling 2.6 1T ling-2.6-1t	InclusionAI	open	31.1%
#64	Qwen3.5 122B qwen3.5-122b-a10b	Alibaba Cloud	open	31.1%
#65	DeepSeek V3.1 Terminus deepseek-v3.1-terminus	DeepSeek	open	30.3%
#66	GPT-5 mini gpt-5-mini	OpenAI	closed	28.8%
#67	MiniMax M2.1 minimax-m2.1	MiniMax	open	28.8%
#68	Nemotron 3 Super nemotron-3-super-120b	NVIDIA	open	28.8%
#69	MiMo-V2-Flash mimo-v2-flash	Xiaomi	open	28.0%
#70	Claude Haiku 4.5 claude-haiku-4-5-20251001	Anthropic	closed	27.3%
#71	Gemini 2.5 Pro gemini-2.5-pro	Google	closed	26.5%
#72	Qwen3.5 35B qwen3.5-35b-a3b	Alibaba Cloud	open	26.5%
#73	MiniMax M2 minimax-m2	MiniMax	open	25.8%
#74	Command A+ command-a-plus	Cohere	open	25%
#75	GLM-4.6 glm-4.6	Z.AI	open	25%
#76	Gemini 3.1 Flash-Lite gemini-3.1-flash-lite	Google	closed	24.2%
#77	Nova 2 Pro nova-2-pro-preview	Amazon	closed	24.2%
#78	Qwen3 Max qwen3-max	Alibaba Cloud	closed	24.2%
#79	Kimi K2 0905 kimi-k2-0905	Moonshot AI	open	23.5%
#80	Claude Sonnet 3.7 claude-3-7-sonnet	Anthropic	closed	21.2%
#81	Qwen3.5 Omni Plus qwen3.5-omni-plus	Alibaba Cloud	closed	21.2%
#82	Grok 4 Fast grok-4-fast	xAI	closed	18.9%
#83	Gemma 4 12B gemma-4-12b	Google	open	18.2%
#84	Grok 3 mini grok-3-mini	xAI	closed	17.4%
#85	Grok Code Fast 1 grok-code-fast-1	xAI	closed	17.4%
#86	Mistral Small 4 mistral-small-2603	Mistral AI	open	17.4%
#87	Gemini 2.5 Flash gemini-2.5-flash	Google	closed	16.7%
#88	Mistral Large 3 mistral-large-2512	Mistral AI	open	15.9%
#89	o4-mini o4-mini	OpenAI	closed	15.2%
#90	Gemma 4 26B gemma-4-26b-a4b	Google	open	13.6%
#91	o1 o1	OpenAI	closed	12.9%

The top of this snapshot is led by Kimi K3 at 85%; third place is 22.1 points behind. The top-10 spread is 32.0 points.

Compare top models

Kimi K3 vs GPT-5.6 Sol GPT-5.6 Sol vs Claude Fable 5 Claude Fable 5 vs GPT-5.5 GPT-5.5 vs Claude Opus 4.8

About Terminal-Bench Hard

Definition and scoring

Organisation: Laude Institute / Artificial Analysis
Category: Agents
Version: Not evaluated
Direction: higher is better
Ranking use: Weighted 1.4%
Contamination risk: Low

Harder agentic terminal evaluation suite spanning software engineering, systems administration, and data processing tasks. Results stay tied to the exact model variant and evaluation system. Multiple systems for the same model use the best published score on this page; overall Lumina scoring uses the median of ranking-eligible rows.

Open benchmark source ↗

FAQ

Quick answers

What does Terminal-Bench Hard measure?

Harder agentic terminal evaluation suite spanning software engineering, systems administration, and data processing tasks.

Which model scores highest on Terminal-Bench Hard?

Kimi K3 by Moonshot AI currently leads with 85%.

How many models are evaluated on Terminal-Bench Hard?

91 models in the LuminaBench cohort have a qualifying score on this benchmark.

Does this affect overall Lumina rank?

Yes. This family is ranking-weighted at 1.4% of its capability category. Kimi K3 is overall #3.

Terminal-Bench

2.1 · 80 results · Weighted 1.6%

Berkeley Function-Calling Leaderboard

V3 · 2 results · Not Weighted

τ-bench

Rolling · 37 results · Weighted 1.6%

GAIA

Rolling · 13 results · Weighted 1.6%

BrowseComp

Rolling · 24 results · Weighted 1.6%

OSWorld

Rolling · 11 results · Weighted 1.6%

Agents · Benchmark profile

Terminal-Bench Hard

Harder agentic terminal evaluation suite spanning software engineering, systems administration, and data processing tasks.

Ranking weightedCurrent / rollingActive91 models

Data verified 15 July 2026 · Methodology 1.2.0

Benchmark score on Terminal-Bench Hard

Leaderboard · 91 models

Kimi K3 leads at 85%, followed by GPT-5.6 Sol (65.9%) and Claude Fable 5 (62.9%).

Ranking · 1.4%AgentsCurrent

#1Closed

Kimi K3

Moonshot AI

kimi-k3

85%

Overall #3 · Context 1M tokens

#2Closed

GPT-5.6 Sol

OpenAI

gpt-5.6-sol

65.9%

Overall #4 · Context 1.05M tokens

#3Closed

Claude Fable 5

Anthropic

claude-fable-5

62.9%

Overall #2 · Context 1M tokens

Visual analysis

See the field, not just the table.

Three views of the same sourced leaderboard: model placement, the score curve, and descriptive provider averages. These visuals never alter ranking calculations.

ModelsColours

Moonshot AIOpenAIAnthropicGoogleZ.AIAlibaba CloudDeepSeek

Leaderboard bars

20 ranked models · higher is better

View accessible chart data

Accessible data for Leaderboard bars
Model	Rank	Provider	Score
Kimi K3	#1	Moonshot AI	85%
GPT-5.6 Sol	#2	OpenAI	65.9%
Claude Fable 5	#3	Anthropic	62.9%
GPT-5.5	#4	OpenAI	60.6%
Claude Opus 4.8	#5	Anthropic	58.3%
GPT-5.4	#6	OpenAI	57.6%
GPT-5.6 Terra	#7	OpenAI	57.6%
Gemini 3.1 Pro Preview	#8	Google	53.8%
Claude Sonnet 4.6	#9	Anthropic	53.0%
GPT-5.3-Codex	#10	OpenAI	53.0%
GPT-5.4 mini	#11	OpenAI	52.3%
Claude Opus 4.7	#12	Anthropic	51.5%
GLM-5.2	#13	Z.AI	50.8%
Qwen3.7-Max	#14	Alibaba Cloud	50.8%
Claude Opus 4.5	#15	Anthropic	47.0%
GPT-5.2	#16	OpenAI	47.0%
Qwen3.7-Plus	#17	Alibaba Cloud	47.0%
Claude Opus 4.6	#18	Anthropic	46.2%
DeepSeek V4 Pro	#19	DeepSeek	46.2%
GPT-5.1	#20	OpenAI	45.5%

Score curve

Shows how quickly scores separate down the field

View accessible chart data

Accessible data for Score curve
Model	Rank	Provider	Score
Kimi K3	#1	Moonshot AI	85%
GPT-5.6 Sol	#2	OpenAI	65.9%
Claude Fable 5	#3	Anthropic	62.9%
GPT-5.5	#4	OpenAI	60.6%
Claude Opus 4.8	#5	Anthropic	58.3%
GPT-5.4	#6	OpenAI	57.6%
GPT-5.6 Terra	#7	OpenAI	57.6%
Gemini 3.1 Pro Preview	#8	Google	53.8%
Claude Sonnet 4.6	#9	Anthropic	53.0%
GPT-5.3-Codex	#10	OpenAI	53.0%
GPT-5.4 mini	#11	OpenAI	52.3%
Claude Opus 4.7	#12	Anthropic	51.5%
GLM-5.2	#13	Z.AI	50.8%
Qwen3.7-Max	#14	Alibaba Cloud	50.8%
Claude Opus 4.5	#15	Anthropic	47.0%
GPT-5.2	#16	OpenAI	47.0%
Qwen3.7-Plus	#17	Alibaba Cloud	47.0%
Claude Opus 4.6	#18	Anthropic	46.2%
DeepSeek V4 Pro	#19	DeepSeek	46.2%
GPT-5.1	#20	OpenAI	45.5%

Provider profile

Average within the visible cohort · descriptive only

View accessible chart data

Accessible data for Provider profile
Provider	Models	Average score
Moonshot AI	1	85%
OpenAI	8	54.9%
Google	1	53.8%
Anthropic	6	53.2%
Z.AI	1	50.8%
Alibaba Cloud	2	48.9%
DeepSeek	1	46.2%

Benchmark score table (91 models)

One best score per model · higher is better

Terminal-Bench Hard leaderboard ranked by score
Rank	Model	Provider	License	Score
#1	Kimi K3 kimi-k3	Moonshot AI	closed	85%
#2	GPT-5.6 Sol gpt-5.6-sol	OpenAI	closed	65.9%
#3	Claude Fable 5 claude-fable-5	Anthropic	closed	62.9%
#4	GPT-5.5 gpt-5.5	OpenAI	closed	60.6%
#5	Claude Opus 4.8 claude-opus-4-8	Anthropic	closed	58.3%
#6	GPT-5.4 gpt-5.4	OpenAI	closed	57.6%
#7	GPT-5.6 Terra gpt-5.6-terra	OpenAI	closed	57.6%
#8	Gemini 3.1 Pro Preview gemini-3.1-pro-preview	Google	closed	53.8%
#9	Claude Sonnet 4.6 claude-sonnet-4-6	Anthropic	closed	53.0%
#10	GPT-5.3-Codex gpt-5.3-codex	OpenAI	closed	53.0%
#11	GPT-5.4 mini gpt-5.4-mini	OpenAI	closed	52.3%
#12	Claude Opus 4.7 claude-opus-4-7	Anthropic	closed	51.5%
#13	GLM-5.2 glm-5-2	Z.AI	open	50.8%
#14	Qwen3.7-Max qwen3.7-max	Alibaba Cloud	—	50.8%
#15	Claude Opus 4.5 claude-opus-4-5	Anthropic	closed	47.0%
#16	GPT-5.2 gpt-5.2	OpenAI	closed	47.0%
#17	Qwen3.7-Plus qwen3.7-plus	Alibaba Cloud	—	47.0%
#18	Claude Opus 4.6 claude-opus-4-6	Anthropic	closed	46.2%
#19	DeepSeek V4 Pro deepseek-v4-pro	DeepSeek	open	46.2%
#20	GPT-5.1 gpt.5.1	OpenAI	closed	45.5%
#21	Muse Spark muse-spark	Meta	closed	45.5%
#22	Kimi K2.7 Code kimi-k2.7-code	Moonshot AI	open	44.7%
#23	Kimi K2.6 kimi-k2.6	Moonshot AI	open	43.9%
#24	Qwen3.6 Max qwen3.6-max	Alibaba Cloud	closed	43.9%
#25	Qwen3.6 Plus qwen3-6-plus	Alibaba Cloud	closed	43.9%
#26	GLM-5 glm-5	Z.AI	open	43.2%
#27	GLM-5.1 glm-5-1	Z.AI	open	43.2%
#28	MiMo-V2.5-Pro mimo-v2-5-pro	Xiaomi	closed	43.2%
#29	GPT-5.4 nano gpt.5.4.nano	OpenAI	closed	42.4%
#30	MiniMax M3 MiniMax-M3	MiniMax	—	42.4%
#31	Gemini 3 Pro gemini-3-pro	Google	closed	41.7%
#32	MiMo-V2.5 mimo-v2.5	Xiaomi	open	41.7%
#33	Gemini 3.5 Flash gemini-3.5-flash	Google	closed	40.9%
#34	MiMo-V2-Pro mimo-v2-pro	Xiaomi	closed	40.9%
#35	Qwen3.5 397B qwen3.5-397b-a17b	Alibaba Cloud	open	40.9%
#36	MiniMax M2.7 minimax-m2-7	MiniMax	open	39.4%
#37	Gemini 3 Flash gemini-3-flash	Google	closed	38.6%
#38	GPT-5 Codex gpt-5-codex	OpenAI	closed	37.9%
#39	Grok 4 grok-4	xAI	closed	37.9%
#40	Grok 4.20 grok-4.20	xAI	closed	37.9%
#41	Grok 4.3 grok-4.3	xAI	closed	37.9%
#42	GPT-5.2 Codex gpt-5.2-codex	OpenAI	closed	37.1%
#43	o3 o3	OpenAI	closed	37.1%
#44	Gemma 4 31B gemma-4-31b	Google	open	36.4%
#45	Nemotron 3 Ultra nemotron-3-ultra-550b	NVIDIA	open	36.4%
#46	Claude Sonnet 4.5 claude-sonnet-4-5	Anthropic	closed	35.6%
#47	DeepSeek V3.2 deepseek-v3.2	DeepSeek	open	35.6%
#48	DeepSeek V4 Flash deepseek-v4-flash	DeepSeek	open	35.6%
#49	Kimi K2.5 kimi-k2.5	Moonshot AI	—	34.8%
#50	MiMo-V2-Omni mimo-v2-omni	Xiaomi	closed	34.8%
#51	MiniMax M2.5 minimax-m2.5	MiniMax	open	34.8%
#52	Qwen3.6 27B qwen3.6-27b	Alibaba Cloud	open	34.8%
#53	Claude Opus 4.1 claude-opus-4-1	Anthropic	closed	34.3%
#54	GLM-5-Turbo glm-5-turbo	Z.AI	closed	33.3%
#55	Mistral Medium 3.5 mistral-medium-3-5	Mistral AI	open	33.3%
#56	GLM-5V-Turbo glm-5v-turbo	Z.AI	closed	32.6%
#57	GPT-5 gpt-5	OpenAI	closed	32.6%
#58	Qwen3.5 27B qwen3.5-27b	Alibaba Cloud	open	32.6%
#59	GLM-4.7 glm-4.7	Z.AI	open	31.8%
#60	Claude Opus 4 claude-opus-4	Anthropic	closed	31.1%
#61	Claude Sonnet 4 claude-sonnet-4	Anthropic	closed	31.1%
#62	Kimi K2 Thinking kimi-k2-thinking	Moonshot AI	open	31.1%
#63	Ling 2.6 1T ling-2.6-1t	InclusionAI	open	31.1%
#64	Qwen3.5 122B qwen3.5-122b-a10b	Alibaba Cloud	open	31.1%
#65	DeepSeek V3.1 Terminus deepseek-v3.1-terminus	DeepSeek	open	30.3%
#66	GPT-5 mini gpt-5-mini	OpenAI	closed	28.8%
#67	MiniMax M2.1 minimax-m2.1	MiniMax	open	28.8%
#68	Nemotron 3 Super nemotron-3-super-120b	NVIDIA	open	28.8%
#69	MiMo-V2-Flash mimo-v2-flash	Xiaomi	open	28.0%
#70	Claude Haiku 4.5 claude-haiku-4-5-20251001	Anthropic	closed	27.3%
#71	Gemini 2.5 Pro gemini-2.5-pro	Google	closed	26.5%
#72	Qwen3.5 35B qwen3.5-35b-a3b	Alibaba Cloud	open	26.5%
#73	MiniMax M2 minimax-m2	MiniMax	open	25.8%
#74	Command A+ command-a-plus	Cohere	open	25%
#75	GLM-4.6 glm-4.6	Z.AI	open	25%
#76	Gemini 3.1 Flash-Lite gemini-3.1-flash-lite	Google	closed	24.2%
#77	Nova 2 Pro nova-2-pro-preview	Amazon	closed	24.2%
#78	Qwen3 Max qwen3-max	Alibaba Cloud	closed	24.2%
#79	Kimi K2 0905 kimi-k2-0905	Moonshot AI	open	23.5%
#80	Claude Sonnet 3.7 claude-3-7-sonnet	Anthropic	closed	21.2%
#81	Qwen3.5 Omni Plus qwen3.5-omni-plus	Alibaba Cloud	closed	21.2%
#82	Grok 4 Fast grok-4-fast	xAI	closed	18.9%
#83	Gemma 4 12B gemma-4-12b	Google	open	18.2%
#84	Grok 3 mini grok-3-mini	xAI	closed	17.4%
#85	Grok Code Fast 1 grok-code-fast-1	xAI	closed	17.4%
#86	Mistral Small 4 mistral-small-2603	Mistral AI	open	17.4%
#87	Gemini 2.5 Flash gemini-2.5-flash	Google	closed	16.7%
#88	Mistral Large 3 mistral-large-2512	Mistral AI	open	15.9%
#89	o4-mini o4-mini	OpenAI	closed	15.2%
#90	Gemma 4 26B gemma-4-26b-a4b	Google	open	13.6%
#91	o1 o1	OpenAI	closed	12.9%

The top of this snapshot is led by Kimi K3 at 85%; third place is 22.1 points behind. The top-10 spread is 32.0 points.

Compare top models

Kimi K3 vs GPT-5.6 Sol GPT-5.6 Sol vs Claude Fable 5 Claude Fable 5 vs GPT-5.5 GPT-5.5 vs Claude Opus 4.8

About Terminal-Bench Hard

Definition and scoring

Organisation: Laude Institute / Artificial Analysis
Category: Agents
Version: Not evaluated
Direction: higher is better
Ranking use: Weighted 1.4%
Contamination risk: Low

Open benchmark source ↗

FAQ

Quick answers

What does Terminal-Bench Hard measure?

Harder agentic terminal evaluation suite spanning software engineering, systems administration, and data processing tasks.

Which model scores highest on Terminal-Bench Hard?

Kimi K3 by Moonshot AI currently leads with 85%.

How many models are evaluated on Terminal-Bench Hard?

91 models in the LuminaBench cohort have a qualifying score on this benchmark.

Does this affect overall Lumina rank?

Yes. This family is ranking-weighted at 1.4% of its capability category. Kimi K3 is overall #3.

Terminal-Bench

2.1 · 80 results · Weighted 1.6%

Berkeley Function-Calling Leaderboard

V3 · 2 results · Not Weighted

τ-bench

Rolling · 37 results · Weighted 1.6%

GAIA

Rolling · 13 results · Weighted 1.6%

BrowseComp

Rolling · 24 results · Weighted 1.6%

OSWorld

Rolling · 11 results · Weighted 1.6%

Leaderboard · 91 models

Kimi K3

GPT-5.6 Sol

Claude Fable 5

See the field, not just the table.

Benchmark score table (91 models)

Compare top models

Definition and scoring

Quick answers

What does Terminal-Bench Hard measure?

Which model scores highest on Terminal-Bench Hard?

How many models are evaluated on Terminal-Bench Hard?

Does this affect overall Lumina rank?

More Agents benchmarks

Leaderboard · 91 models

Kimi K3

GPT-5.6 Sol

Claude Fable 5

See the field, not just the table.

Benchmark score table (91 models)

Compare top models

Definition and scoring

Quick answers

What does Terminal-Bench Hard measure?

Which model scores highest on Terminal-Bench Hard?

How many models are evaluated on Terminal-Bench Hard?

Does this affect overall Lumina rank?

More Agents benchmarks