BenchmarkActivecategory · agenticfamily · terminal-bench · variant 1.0

Terminal-Bench

tbench.ai

terminal tasks solved by agents

quality57

Updated 13 min ago · first seen 11 Sept 2026

Metric: accuracy · % ↑
Current results: 821
Models: 374
Current leader: gpt-5.6-sol 65.9%

Score history · grok-4.20-0309-non-reasoning 1 row

Not enough history to chart — a single observation (16.67% on 11 Sept 2026). Rows under different configurations count separately; the list below shows each one.

16.67%aa_slug=grok-4-20-non-reasoning · variant=hard · evaluator=Artificial Analysis · index_version=4.311 Sept 2026

Back to the leaderboard

Frontier over time · accuracy · variant=hard · evaluator=Artificial Analysis

10 leader changes recorded, all dated 11 Sept 2026 — the frontier line needs at least two distinct dates. The corpus is young: every result was first observed on the same day, so leader changes will separate in time as sources are re-crawled.

65.9%gpt-5.6-sol OpenAI Independent11 Sept 2026
61.4%gpt-5.6-sol OpenAI Independent11 Sept 2026
53.0%Claude Sonnet 4.6 Anthropic Independent11 Sept 2026
51.5%Claude Opus 4.7 Anthropic Independent11 Sept 2026
50.8%Z.ai GLM 5.2 Z.ai (Zhipu AI) Independent11 Sept 2026
49.2%KAT-Coder-Pro V2 Kwaipilot Independent11 Sept 2026
33.3%GPT-5.1-Codex Mini OpenAI Independent11 Sept 2026
26.5%Grok 4.3 xAI Independent11 Sept 2026

Includes closed rows (history). A point is emitted whenever a result beats every earlier result of the same group, ordered by evaluated_at when the source publishes it, else observed_at.

Leaderboard 315 models

Select models with +, then Compare.

No result in this group with these filters

Relax the trust / organization filters or pick another comparability group.

Definition

Task: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 10 h agomedium
Metric: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 10 h agomedium
Metric range: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 13 min agomedium
Variant: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 13 min agomedium
Version: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 13 min agomedium
Harness: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 13 min agomedium
Comparability note: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 13 min agomedium
Website: tbench.ai
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 10 h agomedium

Active group

accuracy · variant=hard · evaluator=Artificial Analysis

315 models · 432 current rows

Comparability

Comparable same task and conditions · Partially comparable same task, conditions differ (effort, temperature, judge) · Not comparable different variant or metric

Also known as

Terminal-Bench, Terminal-Bench 1.0

slug terminal-bench · bench_01M293SPFPKK4MF90JEDK0PH8C

Rows are grouped by comparability group = canonical metric × config_key (hash of the task-defining configuration keys: variant, evaluator, harness, shots, pass regime…). A leaderboard shows one row per canonical model: its best current row inside the group. Reasoning effort, temperature or judge differences keep rows in the same group but mark them partially comparable. delta_rank compares with the ranking built from the closed (previous) rows of the same group.