Terminal task execution on Terminal-Bench 2.0 (full)
58.9Overall avg@5 AccuracySkillsVote
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SkillsVoteBackbone=GPT-5.2, Setting=Offline, Framework=Codex2026.05 | 58.9 | 90 | 65.1 | 43.3 | |
| SkillsVoteBackbone=GPT-5.4 mini, Setting=Offline, Framework=Codex2026.05 | 57.5 | 65 | 64.7 | 43.3 | |
| SkillsVoteBackbone=GPT-5.2, Setting=Online, Framework=Codex2026.05 | 53.7 | 75 | 62.9 | 34 | |
| SkillsVoteBackbone=GPT-5.4 mini, Setting=Online, Framework=Codex2026.05 | 52.8 | 75 | 63.6 | 30 | |
| GPT-5.4 miniSetting=Medium (No-skill baseline), Framework=Codex2026.05 | 51.7 | 75 | 61.8 | 30 | |
| GPT-5.2Setting=Medium (No-skill baseline), Framework=Codex2026.05 | 51 | 75 | 54.9 | 40.7 |