Terminal-Bench 4.0 evaluates complete model-and-agent systems on 66 containerized professional computer tasks. The release gives every agent run an eight-hour timeout, fixes 19 tasks, and removes eight tasks affected by saturation, refusals, public solutions, or unresolved quality and platform issues. Each task is run five times, and the leaderboard reports resolution rate with a 95% confidence interval. Because version 4.0 changes both the task set and resource rules, its scores are not directly comparable with Terminal-Bench 3.0 or 2.x.
Browse the latest scores, model modes, release dates, and parameter sizes for Terminal-Bench 4.0.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
| Rank | Model | License | |||
|---|---|---|---|---|---|
 Claude Opus 5 Thinking Level · MaxTools | 51.82 | 2026-07-24 | Unknown | Closed | |
Claude Fable 5 Thinking Level · MaxTools | 44.55 | 2026-06-09 | Unknown | Closed | |
 GLM-5.3 Thinking Level · MaxTools | 41.82 | 2026-08-14 | 744B | — | |
4 |  GPT-5.6 Sol Thinking Level · MaxTools | 37.27 | 2026-06-26 | Unknown | Closed |
5 | Claude Opus 4.8 Thinking Level · MaxTools | 23.64 | 2026-05-28 | Unknown | Closed |
6 | GPT-5.6 Terra Thinking Level · MaxTools | 21.52 | 2026-06-26 | Unknown | Closed |
7 |  Grok 4.6 Thinking Level · HighTools | 20.30 | 2026-08-12 | Unknown | Closed |
8 | GPT-5.6 Luna Thinking Level · MaxTools | 17.27 | 2026-06-26 | Unknown | Closed |
9 | Claude Sonnet 5 Thinking Level · MaxTools | 12.42 | 2026-06-30 | Unknown | Closed |
10 | Grok 4.5 Thinking Level · HighTools | 12.42 | 2026-07-08 | Unknown | Closed |
In the official leaderboard snapshot reviewed on August 30, 2026, Claude Opus 5 with Claude Code at max reasoning leads at 51.82%, followed by Claude Fable 5 with Claude Code at 44.55% and GLM-5.3 with Claude Code at 41.82%. These are model-and-agent system results, not agent-independent model scores.
