Coding
Terminal-Bench 2.0
The 2.0 task set of the terminal agent benchmark — the version most launches reported from late 2025 to spring 2026. Not comparable with 2.1 or 4.0.
Terminal-Bench 2.0 is scored as a percentage of questions answered correctly. 5 of the 189 models in this index report a score for it. The highest published figure here is 82.7%, from GPT-5.5.
Reported scores
Top 5 on Terminal-Bench 2.0
Ordered by the figure each maker published. Models that have not reported this benchmark are not listed — an absent score is not a low score.
| # | Model | Published score |
|---|---|---|
| 1 | GPT-5.5OpenAI | |
| 2 | Gemini 3.1 ProGoogle DeepMind | |
| 3 | Claude Opus 4.6Anthropic | |
| 4 | Claude Sonnet 4.6Anthropic | |
| 5 | GLM-5Z.ai (Zhipu) |
Where these numbers come from
Every score on this page is a published figure, taken from the model's own card, system card, technical report or release post, or from a public leaderboard. CorX Labs did not run these evaluations. Most are self-reported by the lab that built the model, which means they were produced under that lab's own choice of prompt, scaffold and number of attempts — so treat them as a starting point for a shortlist, not as a settled ranking.
A score someone other than the model's maker measured is marked Independent and names its measurer. Those are the stronger numbers on this page — an outside harness has no reason to flatter anyone — and there are not many of them.
Where a figure has not been published, the cell reads Not reported rather than an estimate. Nothing here is inferred, interpolated or guessed. Each model records the month its row was last checked. Full method and caveats.
How to read this score
The 2.0 task set of Terminal-Bench, and the version almost every launch from late 2025 to spring 2026 reported. It is the only terminal figure most models from that window have, which is why it gets its own column.
Where it is weak
Do not read it against 2.1 or 4.0 — the task sets differ. Labs also ran it on very different harnesses, from the public Terminus-2 agent to their own coding CLIs, and the harness moves the score as much as the model does.