Agents · Benchmark profile

Long-Horizon-Terminal-Bench

A 46-task terminal benchmark for long-running agent work with dense subtask grading across nine practical domains.

Ranking weighted2026-07Active37 models

Data verified 15 July 2026 · Methodology 1.2.0

Benchmark score on Long-Horizon-Terminal-Bench

Leaderboard · 37 models

GPT-5.6 Sol leads at 65.9%, followed by Claude Fable 5 (62.9%) and GPT-5.5 (60.6%).

Ranking · 1.6%Agents2026-07

#1Closed

GPT-5.6 Sol

OpenAI

gpt-5.6-sol

65.9%

Overall #4 · Context 1.05M tokens

#2Closed

Claude Fable 5

Anthropic

claude-fable-5

62.9%

Overall #2 · Context 1M tokens

#3Closed

GPT-5.5

OpenAI

gpt-5.5

60.6%

Overall #8 · Context 1.05M 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

OpenAIAnthropicGoogleZ.AIAlibaba CloudMetaXiaomi

Leaderboard bars

20 ranked models · higher is better

View accessible chart data

Accessible data for Leaderboard bars
Model	Rank	Provider	Score
GPT-5.6 Sol	#1	OpenAI	65.9%
Claude Fable 5	#2	Anthropic	62.9%
GPT-5.5	#3	OpenAI	60.6%
Claude Opus 4.8	#4	Anthropic	58.3%
GPT-5.4	#5	OpenAI	57.6%
GPT-5.6 Terra	#6	OpenAI	57.6%
Claude Opus 4.7	#7	Anthropic	54.5%
Gemini 3.1 Pro Preview	#8	Google	53.8%
GPT-5.3-Codex	#9	OpenAI	53%
GLM-5.2	#10	Z.AI	50.8%
Qwen3.7-Max	#11	Alibaba Cloud	50.8%
Claude Opus 4.6	#12	Anthropic	48.5%
Qwen3.7-Plus	#13	Alibaba Cloud	47%
Claude Sonnet 4.6	#14	Anthropic	46.2%
GPT-5.1	#15	OpenAI	45.5%
Muse Spark	#16	Meta	45.5%
Qwen3.6 Plus	#17	Alibaba Cloud	43.9%
GLM-5	#18	Z.AI	43.2%
GLM-5.1	#19	Z.AI	43.2%
MiMo-V2.5-Pro	#20	Xiaomi	43.2%

Score curve

Shows how quickly scores separate down the field

View accessible chart data

Accessible data for Score curve
Model	Rank	Provider	Score
GPT-5.6 Sol	#1	OpenAI	65.9%
Claude Fable 5	#2	Anthropic	62.9%
GPT-5.5	#3	OpenAI	60.6%
Claude Opus 4.8	#4	Anthropic	58.3%
GPT-5.4	#5	OpenAI	57.6%
GPT-5.6 Terra	#6	OpenAI	57.6%
Claude Opus 4.7	#7	Anthropic	54.5%
Gemini 3.1 Pro Preview	#8	Google	53.8%
GPT-5.3-Codex	#9	OpenAI	53%
GLM-5.2	#10	Z.AI	50.8%
Qwen3.7-Max	#11	Alibaba Cloud	50.8%
Claude Opus 4.6	#12	Anthropic	48.5%
Qwen3.7-Plus	#13	Alibaba Cloud	47%
Claude Sonnet 4.6	#14	Anthropic	46.2%
GPT-5.1	#15	OpenAI	45.5%
Muse Spark	#16	Meta	45.5%
Qwen3.6 Plus	#17	Alibaba Cloud	43.9%
GLM-5	#18	Z.AI	43.2%
GLM-5.1	#19	Z.AI	43.2%
MiMo-V2.5-Pro	#20	Xiaomi	43.2%

Provider profile

Average within the visible cohort · descriptive only

View accessible chart data

Accessible data for Provider profile
Provider	Models	Average score
OpenAI	6	56.7%
Anthropic	5	54.1%
Google	1	53.8%
Alibaba Cloud	3	47.2%
Z.AI	3	45.7%
Meta	1	45.5%
Xiaomi	1	43.2%

Benchmark score table (37 models)

One best score per model · higher is better

Long-Horizon-Terminal-Bench leaderboard ranked by score
Rank	Model	Provider	License	Score
#1	GPT-5.6 Sol gpt-5.6-sol	OpenAI	closed	65.9%
#2	Claude Fable 5 claude-fable-5	Anthropic	closed	62.9%
#3	GPT-5.5 gpt-5.5	OpenAI	closed	60.6%
#4	Claude Opus 4.8 claude-opus-4-8	Anthropic	closed	58.3%
#5	GPT-5.4 gpt-5.4	OpenAI	closed	57.6%
#6	GPT-5.6 Terra gpt-5.6-terra	OpenAI	closed	57.6%
#7	Claude Opus 4.7 claude-opus-4-7	Anthropic	closed	54.5%
#8	Gemini 3.1 Pro Preview gemini-3.1-pro-preview	Google	closed	53.8%
#9	GPT-5.3-Codex gpt-5.3-codex	OpenAI	closed	53%
#10	GLM-5.2 glm-5-2	Z.AI	open	50.8%
#11	Qwen3.7-Max qwen3.7-max	Alibaba Cloud	—	50.8%
#12	Claude Opus 4.6 claude-opus-4-6	Anthropic	closed	48.5%
#13	Qwen3.7-Plus qwen3.7-plus	Alibaba Cloud	—	47%
#14	Claude Sonnet 4.6 claude-sonnet-4-6	Anthropic	closed	46.2%
#15	GPT-5.1 gpt.5.1	OpenAI	closed	45.5%
#16	Muse Spark muse-spark	Meta	closed	45.5%
#17	Qwen3.6 Plus qwen3-6-plus	Alibaba Cloud	closed	43.9%
#18	GLM-5 glm-5	Z.AI	open	43.2%
#19	GLM-5.1 glm-5-1	Z.AI	open	43.2%
#20	MiMo-V2.5-Pro mimo-v2-5-pro	Xiaomi	closed	43.2%
#21	GPT-5.4 nano gpt.5.4.nano	OpenAI	closed	42.4%
#22	MiniMax M3 MiniMax-M3	MiniMax	—	42.4%
#23	Gemini 3 Pro gemini-3-pro	Google	closed	41.7%
#24	Claude Opus 4.5 claude-opus-4-5	Anthropic	closed	40.9%
#25	Gemini 3.5 Flash gemini-3.5-flash	Google	closed	40.9%
#26	MiMo-V2-Pro mimo-v2-pro	Xiaomi	closed	40.9%
#27	MiniMax M2.7 minimax-m2-7	MiniMax	open	39.4%
#28	Grok 4.3 grok-4.3	xAI	closed	37.9%
#29	Gemma 4 31B gemma-4-31b	Google	open	36.4%
#30	Kimi K2.5 kimi-k2.5	Moonshot AI	—	34.8%
#31	MiMo-V2-Omni mimo-v2-omni	Xiaomi	closed	34.8%
#32	GLM-5-Turbo glm-5-turbo	Z.AI	closed	33.3%
#33	GLM-5V-Turbo glm-5v-turbo	Z.AI	closed	32.6%
#34	Gemini 3 Flash gemini-3-flash	Google	closed	31.8%
#35	Gemini 3.1 Flash-Lite gemini-3.1-flash-lite	Google	closed	24.2%
#36	Mistral Small 4 mistral-small-2603	Mistral AI	open	17.4%
#37	Mistral Large 3 mistral-large-2512	Mistral AI	open	15.9%

The top of this snapshot is led by GPT-5.6 Sol at 65.9%; third place is 5.3 points behind. The top-10 spread is 15.1 points.

Compare top models

GPT-5.6 Sol vs Claude Fable 5 Claude Fable 5 vs GPT-5.5 GPT-5.5 vs Claude Opus 4.8 Claude Opus 4.8 vs GPT-5.4

About Long-Horizon-Terminal-Bench

Definition and scoring

Organisation: Long-Horizon-Terminal-Bench authors
Category: Agents
Version: 2026-07
Direction: higher is better
Ranking use: Weighted 1.6%
Contamination risk: Low

A 46-task terminal benchmark for long-running agent work with dense subtask grading across nine practical domains. 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 Long-Horizon-Terminal-Bench measure?

A 46-task terminal benchmark for long-running agent work with dense subtask grading across nine practical domains.

Which model scores highest on Long-Horizon-Terminal-Bench?

GPT-5.6 Sol by OpenAI currently leads with 65.9%.

How many models are evaluated on Long-Horizon-Terminal-Bench?

37 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 1.6% of its capability category. GPT-5.6 Sol is overall #4.

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

Long-Horizon-Terminal-Bench

A 46-task terminal benchmark for long-running agent work with dense subtask grading across nine practical domains.

Ranking weighted2026-07Active37 models

Data verified 15 July 2026 · Methodology 1.2.0

Benchmark score on Long-Horizon-Terminal-Bench

Leaderboard · 37 models

GPT-5.6 Sol leads at 65.9%, followed by Claude Fable 5 (62.9%) and GPT-5.5 (60.6%).

Ranking · 1.6%Agents2026-07

#1Closed

GPT-5.6 Sol

OpenAI

gpt-5.6-sol

65.9%

Overall #4 · Context 1.05M tokens

#2Closed

Claude Fable 5

Anthropic

claude-fable-5

62.9%

Overall #2 · Context 1M tokens

#3Closed

GPT-5.5

OpenAI

gpt-5.5

60.6%

Overall #8 · Context 1.05M 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

OpenAIAnthropicGoogleZ.AIAlibaba CloudMetaXiaomi

Leaderboard bars

20 ranked models · higher is better

View accessible chart data

Accessible data for Leaderboard bars
Model	Rank	Provider	Score
GPT-5.6 Sol	#1	OpenAI	65.9%
Claude Fable 5	#2	Anthropic	62.9%
GPT-5.5	#3	OpenAI	60.6%
Claude Opus 4.8	#4	Anthropic	58.3%
GPT-5.4	#5	OpenAI	57.6%
GPT-5.6 Terra	#6	OpenAI	57.6%
Claude Opus 4.7	#7	Anthropic	54.5%
Gemini 3.1 Pro Preview	#8	Google	53.8%
GPT-5.3-Codex	#9	OpenAI	53%
GLM-5.2	#10	Z.AI	50.8%
Qwen3.7-Max	#11	Alibaba Cloud	50.8%
Claude Opus 4.6	#12	Anthropic	48.5%
Qwen3.7-Plus	#13	Alibaba Cloud	47%
Claude Sonnet 4.6	#14	Anthropic	46.2%
GPT-5.1	#15	OpenAI	45.5%
Muse Spark	#16	Meta	45.5%
Qwen3.6 Plus	#17	Alibaba Cloud	43.9%
GLM-5	#18	Z.AI	43.2%
GLM-5.1	#19	Z.AI	43.2%
MiMo-V2.5-Pro	#20	Xiaomi	43.2%

Score curve

Shows how quickly scores separate down the field

View accessible chart data

Accessible data for Score curve
Model	Rank	Provider	Score
GPT-5.6 Sol	#1	OpenAI	65.9%
Claude Fable 5	#2	Anthropic	62.9%
GPT-5.5	#3	OpenAI	60.6%
Claude Opus 4.8	#4	Anthropic	58.3%
GPT-5.4	#5	OpenAI	57.6%
GPT-5.6 Terra	#6	OpenAI	57.6%
Claude Opus 4.7	#7	Anthropic	54.5%
Gemini 3.1 Pro Preview	#8	Google	53.8%
GPT-5.3-Codex	#9	OpenAI	53%
GLM-5.2	#10	Z.AI	50.8%
Qwen3.7-Max	#11	Alibaba Cloud	50.8%
Claude Opus 4.6	#12	Anthropic	48.5%
Qwen3.7-Plus	#13	Alibaba Cloud	47%
Claude Sonnet 4.6	#14	Anthropic	46.2%
GPT-5.1	#15	OpenAI	45.5%
Muse Spark	#16	Meta	45.5%
Qwen3.6 Plus	#17	Alibaba Cloud	43.9%
GLM-5	#18	Z.AI	43.2%
GLM-5.1	#19	Z.AI	43.2%
MiMo-V2.5-Pro	#20	Xiaomi	43.2%

Provider profile

Average within the visible cohort · descriptive only

View accessible chart data

Accessible data for Provider profile
Provider	Models	Average score
OpenAI	6	56.7%
Anthropic	5	54.1%
Google	1	53.8%
Alibaba Cloud	3	47.2%
Z.AI	3	45.7%
Meta	1	45.5%
Xiaomi	1	43.2%

Benchmark score table (37 models)

One best score per model · higher is better

Long-Horizon-Terminal-Bench leaderboard ranked by score
Rank	Model	Provider	License	Score
#1	GPT-5.6 Sol gpt-5.6-sol	OpenAI	closed	65.9%
#2	Claude Fable 5 claude-fable-5	Anthropic	closed	62.9%
#3	GPT-5.5 gpt-5.5	OpenAI	closed	60.6%
#4	Claude Opus 4.8 claude-opus-4-8	Anthropic	closed	58.3%
#5	GPT-5.4 gpt-5.4	OpenAI	closed	57.6%
#6	GPT-5.6 Terra gpt-5.6-terra	OpenAI	closed	57.6%
#7	Claude Opus 4.7 claude-opus-4-7	Anthropic	closed	54.5%
#8	Gemini 3.1 Pro Preview gemini-3.1-pro-preview	Google	closed	53.8%
#9	GPT-5.3-Codex gpt-5.3-codex	OpenAI	closed	53%
#10	GLM-5.2 glm-5-2	Z.AI	open	50.8%
#11	Qwen3.7-Max qwen3.7-max	Alibaba Cloud	—	50.8%
#12	Claude Opus 4.6 claude-opus-4-6	Anthropic	closed	48.5%
#13	Qwen3.7-Plus qwen3.7-plus	Alibaba Cloud	—	47%
#14	Claude Sonnet 4.6 claude-sonnet-4-6	Anthropic	closed	46.2%
#15	GPT-5.1 gpt.5.1	OpenAI	closed	45.5%
#16	Muse Spark muse-spark	Meta	closed	45.5%
#17	Qwen3.6 Plus qwen3-6-plus	Alibaba Cloud	closed	43.9%
#18	GLM-5 glm-5	Z.AI	open	43.2%
#19	GLM-5.1 glm-5-1	Z.AI	open	43.2%
#20	MiMo-V2.5-Pro mimo-v2-5-pro	Xiaomi	closed	43.2%
#21	GPT-5.4 nano gpt.5.4.nano	OpenAI	closed	42.4%
#22	MiniMax M3 MiniMax-M3	MiniMax	—	42.4%
#23	Gemini 3 Pro gemini-3-pro	Google	closed	41.7%
#24	Claude Opus 4.5 claude-opus-4-5	Anthropic	closed	40.9%
#25	Gemini 3.5 Flash gemini-3.5-flash	Google	closed	40.9%
#26	MiMo-V2-Pro mimo-v2-pro	Xiaomi	closed	40.9%
#27	MiniMax M2.7 minimax-m2-7	MiniMax	open	39.4%
#28	Grok 4.3 grok-4.3	xAI	closed	37.9%
#29	Gemma 4 31B gemma-4-31b	Google	open	36.4%
#30	Kimi K2.5 kimi-k2.5	Moonshot AI	—	34.8%
#31	MiMo-V2-Omni mimo-v2-omni	Xiaomi	closed	34.8%
#32	GLM-5-Turbo glm-5-turbo	Z.AI	closed	33.3%
#33	GLM-5V-Turbo glm-5v-turbo	Z.AI	closed	32.6%
#34	Gemini 3 Flash gemini-3-flash	Google	closed	31.8%
#35	Gemini 3.1 Flash-Lite gemini-3.1-flash-lite	Google	closed	24.2%
#36	Mistral Small 4 mistral-small-2603	Mistral AI	open	17.4%
#37	Mistral Large 3 mistral-large-2512	Mistral AI	open	15.9%

The top of this snapshot is led by GPT-5.6 Sol at 65.9%; third place is 5.3 points behind. The top-10 spread is 15.1 points.

Compare top models

GPT-5.6 Sol vs Claude Fable 5 Claude Fable 5 vs GPT-5.5 GPT-5.5 vs Claude Opus 4.8 Claude Opus 4.8 vs GPT-5.4

About Long-Horizon-Terminal-Bench

Definition and scoring

Organisation: Long-Horizon-Terminal-Bench authors
Category: Agents
Version: 2026-07
Direction: higher is better
Ranking use: Weighted 1.6%
Contamination risk: Low

Open benchmark source ↗

FAQ

Quick answers

What does Long-Horizon-Terminal-Bench measure?

A 46-task terminal benchmark for long-running agent work with dense subtask grading across nine practical domains.

Which model scores highest on Long-Horizon-Terminal-Bench?

GPT-5.6 Sol by OpenAI currently leads with 65.9%.

How many models are evaluated on Long-Horizon-Terminal-Bench?

37 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 1.6% of its capability category. GPT-5.6 Sol is overall #4.

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 · 37 models

GPT-5.6 Sol

Claude Fable 5

GPT-5.5

See the field, not just the table.

Benchmark score table (37 models)

Compare top models

Definition and scoring

Quick answers

What does Long-Horizon-Terminal-Bench measure?

Which model scores highest on Long-Horizon-Terminal-Bench?

How many models are evaluated on Long-Horizon-Terminal-Bench?

Does this affect overall Lumina rank?

More Agents benchmarks

Leaderboard · 37 models

GPT-5.6 Sol

Claude Fable 5

GPT-5.5

See the field, not just the table.

Benchmark score table (37 models)

Compare top models

Definition and scoring

Quick answers

What does Long-Horizon-Terminal-Bench measure?

Which model scores highest on Long-Horizon-Terminal-Bench?

How many models are evaluated on Long-Horizon-Terminal-Bench?

Does this affect overall Lumina rank?

More Agents benchmarks