Agents · Benchmark profile

Terminal-Bench 3.0

Version 3.0 of the terminal-agent benchmark, retained as a reference-only family until methodology 1.8 defines a compatible scoring anchor.

Reference3.0Active9 models

Data verified 12 Aug 2026 · Methodology 2.2.0

Benchmark score on Terminal-Bench 3.0

Leaderboard · 9 models

GPT-5.6 Sol leads at 34.6%, followed by Claude Fable 5 (34.1%) and GLM-5.3 (28.3%).

Display onlyAgents3.0

#1Closed

GPT-5.6 Sol

OpenAI

gpt-5.6-sol

34.6%

Overall #3 · Context 1.05M tokens

#2Closed

Claude Fable 5

Anthropic

claude-fable-5

34.1%

Overall #2 · Context 1M tokens

#3Closed

GLM-5.3

Z.AI

glm-5.3

28.3%

Overall — · Context 1M tokens

ViewModelsColours

OpenAIAnthropicZ.AIxAIGoogle

Lumina

Terminal-Bench 3.0 leaderboard

Accessible data for Terminal-Bench 3.0 leaderboard
Model	Rank	Model ID	Provider	Score	Unit
GPT-5.6 Sol	#1	gpt-5-6-sol	OpenAI	34.6	percent
Claude Fable 5	#2	claude-fable-5	Anthropic	34.1	percent
GLM-5.3	#3	glm-5-3	Z.AI	28.3	percent
Grok 4.6	#4	grok-4-6	xAI	26	percent
GPT-5.6 Terra	#5	gpt-5-6-terra	OpenAI	20.8	percent
Grok 4.5	#6	grok-4-5	xAI	15.7	percent
Gemini 3.7 Flash	#7	gemini-3-7-flash	Google	14.9	percent
Claude Sonnet 5	#8	claude-sonnet-5	Anthropic	14.6	percent
Gemini 3.6 Flash	#9	gemini-3-6-flash	Google	5.4	percent

Benchmark score table (9 models)

One best compatible score per canonical product · higher is better

Terminal-Bench 3.0 leaderboard ranked by score
Rank	Model	Provider	License	Evidence use	Score
#1	GPT-5.6 Sol gpt-5.6-sol	OpenAI	closed	Reference only	34.6%
#2	Claude Fable 5 claude-fable-5	Anthropic	closed	Reference only	34.1%
#3	GLM-5.3 glm-5.3	Z.AI	announced-open	Reference only	28.3%
#4	Grok 4.6 grok-4.6	xAI	closed	Reference only	26%
#5	GPT-5.6 Terra gpt-5.6-terra	OpenAI	closed	Reference only	20.8%
#6	Grok 4.5 grok-4.5	xAI	closed	Reference only	15.7%
#7	Gemini 3.7 Flash gemini-3.7-flash	Google	closed	Reference only	14.9%
#8	Claude Sonnet 5 claude-sonnet-5	Anthropic	closed	Reference only	14.6%
#9	Gemini 3.6 Flash gemini-3.6-flash	Google	closed	Reference only	5.4%

The top of this snapshot is led by GPT-5.6 Sol at 34.6%; third place is 6.3 points behind. The top-9 spread is 29.2 points.

Configuration-preserving evidencePublished source rows (14)Exact source labels, configurations, versions and dated observations. Same-snapshot canonical/configuration projections are shown once.

Every distinct dated source-published row for Terminal-Bench 3.0, including exact model configuration labels
Exact published model / configuration	Canonical product	Score	Version / system	Evidence	Source / observed
GPT-5.6 Sol as published by Z.AI gpt-5-6-sol-max	GPT-5.6 Sol	34.6%	3.0Z.AI GLM-5.3 official comparison table	Excluded provider-reported	GLM-5.3: Frontier Coding with Emergent Cyber Capabilities ↗Observed 2026-08-14 · checked 2026-08-15
Claude Fable 5 w/ fallback as published by Z.AI claude-fable-5-max	Claude Fable 5	33.7%	3.0Z.AI GLM-5.3 official comparison table	Excluded provider-reported	GLM-5.3: Frontier Coding with Emergent Cyber Capabilities ↗Observed 2026-08-14 · checked 2026-08-15
GLM-5.3 (max) glm-5-3-max	GLM-5.3	28.3%	3.0Z.AI GLM-5.3 official comparison table	Reference only provider-reported	GLM-5.3: Frontier Coding with Emergent Cyber Capabilities ↗Observed 2026-08-14 · checked 2026-08-15
Claude Opus 4.8 as published by Z.AI claude-opus-4-8-max	Claude Opus 4.8	21.1%	3.0Z.AI GLM-5.3 official comparison table	Excluded provider-reported	GLM-5.3: Frontier Coding with Emergent Cyber Capabilities ↗Observed 2026-08-14 · checked 2026-08-15
Kimi K3 as published by Z.AI kimi-k3-max	Kimi K3	17.4%	3.0Z.AI GLM-5.3 official comparison table	Excluded provider-reported	GLM-5.3: Frontier Coding with Emergent Cyber Capabilities ↗Observed 2026-08-14 · checked 2026-08-15
GLM-5.2 as published by Z.AI glm-5-2-max	GLM-5.2	4.6%	3.0Z.AI GLM-5.3 official comparison table	Excluded provider-reported	GLM-5.3: Frontier Coding with Emergent Cyber Capabilities ↗Observed 2026-08-14 · checked 2026-08-15
GPT-5.6 Terra gpt-5-6-terra-max	GPT-5.6 Terra	20.8%	3.0Terminal-Bench 3.0; Gemini self-run used mini SWE agent and LiteLLM 1.96	Reference only provider-reported	Google DeepMind permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
Gemini 3.7 Flash gemini-3-7-flash-high	Gemini 3.7 Flash	14.9%	3.0Terminal-Bench 3.0; Gemini self-run used mini SWE agent and LiteLLM 1.96	Reference only provider-reported	Google DeepMind permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
Claude Sonnet 5 claude-sonnet-5-max	Claude Sonnet 5	14.6%	3.0Terminal-Bench 3.0; Gemini self-run used mini SWE agent and LiteLLM 1.96	Reference only provider-reported	Google DeepMind permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
Gemini 3.6 Flash gemini-3-6-flash-high	Gemini 3.6 Flash	5.4%	3.0Terminal-Bench 3.0; Gemini self-run used mini SWE agent and LiteLLM 1.96	Reference only provider-reported	Google DeepMind permanent refresh source ↗Observed 2026-08-13 · checked 2026-08-13
GPT-5.6 Sol Max gpt-5-6-sol-max	GPT-5.6 Sol	34.6%	3.0xAI Grok 4.6 official release comparison table	Reference only provider-reported	Grok 4.6 ↗Observed 2026-08-12 · checked 2026-08-12
Fable 5 Max claude-fable-5-max	Claude Fable 5	34.1%	3.0xAI Grok 4.6 official release comparison table	Reference only provider-reported	Grok 4.6 ↗Observed 2026-08-12 · checked 2026-08-12
Grok 4.6 High grok-4-6-high	Grok 4.6	26%	3.0xAI Grok 4.6 official release comparison table	Reference only provider-reported	Grok 4.6 ↗Observed 2026-08-12 · checked 2026-08-12
Grok 4.5 High grok-4-5-aa-2-high	Grok 4.5	15.7%	3.0xAI Grok 4.6 official release comparison table	Reference only provider-reported	Grok 4.6 ↗Observed 2026-08-12 · checked 2026-08-12

Compare top models

GPT-5.6 Sol vs Claude Fable 5 Claude Fable 5 vs GLM-5.3 GLM-5.3 vs Grok 4.6 Grok 4.6 vs GPT-5.6 Terra

About Terminal-Bench 3.0

Definition and scoring

Organisation: Harbor / Terminal-Bench
Category: Agents
Version: 3.0
Direction: higher is better
Ranking use: Reference
Contamination risk: Unknown

Version 3.0 of the terminal-agent benchmark, retained as a reference-only family until methodology 1.8 defines a compatible scoring anchor. Every genuine source row stays tied to its exact model label, configuration, benchmark version and evaluation system. The summary chart shows one best compatible score per canonical product; Score 2.0 admits only explicitly mapped, frozen protocols and keeps incompatible configurations separate.

Open benchmark source ↗

FAQ

Quick answers

What does Terminal-Bench 3.0 measure?

Version 3.0 of the terminal-agent benchmark, retained as a reference-only family until methodology 1.8 defines a compatible scoring anchor.

Which model scores highest on Terminal-Bench 3.0?

GPT-5.6 Sol by OpenAI currently leads with 34.6%.

How many models are evaluated on Terminal-Bench 3.0?

9 canonical model products have a qualifying score, backed by 14 distinct dated source rows.

Does this affect overall Lumina rank?

No. This benchmark is display-only and does not enter the overall Lumina composite.

Agents · Benchmark profile

Terminal-Bench 3.0

Version 3.0 of the terminal-agent benchmark, retained as a reference-only family until methodology 1.8 defines a compatible scoring anchor.

Reference3.0Active9 models

Data verified 12 Aug 2026 · Methodology 2.2.0

Benchmark score on Terminal-Bench 3.0