Best AI for AI agents and terminal automation (Oct 2026)
As of 19 Oct 2026, the highest reported score on Terminal-Bench 4.0 is Claude Opus 5.5 (Anthropic) at 65.15%, per Vals AI. As of 28 Sep 2026, the highest reported score on Terminal-Bench 2.1 is Claude Opus 5.5 (Anthropic) at 87.64%, per Vals AI. The source no longer updates this leaderboard.
Results from published AI leaderboards. Compare models within each benchmark; scores from different tests aren’t directly comparable. How we collect these results
Terminal-Bench 4.0
Frontier-difficulty terminal tasks across software, science, ML, operations, hardware, security and media. Run by Vals AI (benchmark by Terminal-Bench team). Leaderboard dated 19 Oct 2026. Last checked 21 Oct 2026.
- Anthropic
- OpenAI
- xAI
- Meta
- DeepSeek
- Alibaba
- Z.ai
- MiniMax
- Mistral
- Xiaomi
- Other
Full Terminal-Bench 4.0 leaderboard: every model, all metrics and the table →
Terminal-Bench 2.1
A set of difficult terminal-based tasks, now superseded by Terminal-Bench 4.0. Run by Vals AI (benchmark by Terminal-Bench team). Leaderboard dated 28 Sep 2026 (no longer updated by the source). Last checked 21 Oct 2026.
- Anthropic
- OpenAI
- xAI
- Meta
- DeepSeek
- Moonshot
- Xiaomi
Full Terminal-Bench 2.1 leaderboard: every model, all metrics and the table →
What these benchmarks tell you
- Terminal-Bench 4.0: the share of hard, multi-step command-line tasks (software, science, ML, operations, hardware, security, media) an agent completes on its own, verified by tests, plus cost per task.
- Terminal-Bench 2.1: the earlier task set, now superseded and no longer updated by the source.
What they don’t tell you
- It tests agents in a terminal, not agents working inside your business apps, inbox or CRM.
- It does not measure safety, permissions handling or how well an agent asks for help.
Prompts for AI agents and terminal automation
Run the same prompt on your own material with two models and compare the results.
Michael | AI + Agents @Michael_Fenech_Report or remove
Questions people ask
Which AI has the highest reported score for AI agents and terminal automation?
As of 19 Oct 2026, the highest reported score on Terminal-Bench 4.0 is Claude Opus 5.5 (Anthropic) at 65.15%, per Vals AI. As of 28 Sep 2026, the highest reported score on Terminal-Bench 2.1 is Claude Opus 5.5 (Anthropic) at 87.64%, per Vals AI. The source no longer updates this leaderboard.
How recent are these results?
Terminal-Bench 4.0: leaderboard dated 19 Oct 2026; we last checked it on 21 Oct 2026. Terminal-Bench 2.1: leaderboard dated 28 Sep 2026, no longer updated by the source; we last checked it on 21 Oct 2026.
What does the top-scoring model cost to run on these tests?
On Terminal-Bench 4.0, Claude Opus 5.5 costs $13 (cost per test), as published by Vals AI. On Terminal-Bench 2.1, Claude Opus 5.5 costs $0.50 (cost per test), as published by Vals AI. That is API cost on the benchmark’s own tasks, not a subscription price.
Who runs these benchmarks?
Terminal-Bench 4.0 is run by Vals AI (benchmark by Terminal-Bench team). Terminal-Bench 2.1 is run by Vals AI (benchmark by Terminal-Bench team). ChatJudge does not run the tests; we copy the published scores with a link to the source.
What don't these benchmarks tell you about AI agents and terminal automation?
It tests agents in a terminal, not agents working inside your business apps, inbox or CRM. It does not measure safety, permissions handling or how well an agent asks for help.