LLMs TigerLive LLM benchmarks

Terminal-Bench 4.0 leaderboard: which LLMs score highest

Terminal-Bench 4.0 (real terminal tasks, agent+model config) currently has scores for 15 models. The top score is 58.2% by Codex + GPT-6 Astra; the median model scores 23.6% and the lowest scores 11.2%.

Terminal-Bench 4.0 leaderboard: top 15 of 15 models (highest-effort row per model)
#ModelOrganisationTerminal-Bench 4.0 scoreCheapest input $/MSourceCaptured
1Codex + GPT-6 Astra—58.2%—Terminal-Bench 4.0 official leaderboard2026-10-05
2Claude Code + Fable 5.1—57.9%—Terminal-Bench 4.0 official leaderboard2026-10-05
3Claude Code + Opus 5—53.9%—Terminal-Bench 4.0 official leaderboard2026-10-05
4Claude Code + Fable 5—44.5%—Terminal-Bench 4.0 official leaderboard2026-10-05
5Claude Code + GLM-5.3—41.8%—Terminal-Bench 4.0 official leaderboard2026-10-05
6Grok Build + Grok 4.7—37.6%—Terminal-Bench 4.0 official leaderboard2026-10-05
7Codex + GPT-5.6 Sol—37.3%—Terminal-Bench 4.0 official leaderboard2026-10-05
8Claude Code + Opus 4.8—23.6%—Terminal-Bench 4.0 official leaderboard2026-10-05
9Codex + GPT-5.6 Terra—21.5%—Terminal-Bench 4.0 official leaderboard2026-10-05
10Grok Build + Grok 4.6—20.3%—Terminal-Bench 4.0 official leaderboard2026-10-05
11mini-SWE-agent + Gemini 3.8 Flash—19.1%—Terminal-Bench 4.0 official leaderboard2026-10-05
12Codex + GPT-5.6 Luna—17.3%—Terminal-Bench 4.0 official leaderboard2026-10-05
13Claude Code + Sonnet 5—12.4%—Terminal-Bench 4.0 official leaderboard2026-10-05
14Grok Build + Grok 4.5—12.4%—Terminal-Bench 4.0 official leaderboard2026-10-05
15mini-SWE-agent + Gemini 3.7 Flash—11.2%—Terminal-Bench 4.0 official leaderboard2026-10-05

Compare all benchmarks side by side on the live leaderboard →

What Terminal-Bench 4.0 measures

Terminal-Bench tests agents on real tasks in a command-line environment. Scores depend on both the model and the agent harness around it.

How to read these scores

Scores are percentages (higher is better): the share of questions or tasks the model completed correctly. Where a model is listed at several reasoning efforts, only its highest-effort row is shown. See the methodology for how rows are chosen.

Where the data comes from

Terminal-Bench 4.0 official leaderboard. Scores were last captured 2026-10-05; the Captured column gives each row's date.