Terminal-Bench 2.0
- OpenAI
- Anthropic
- Reported in a later post or card
Results
| Model | Score | Source | Human-verified |
|---|---|---|---|
| Claude Opus 4.716 Apr 2026 · Anthropic | 69.4% | Introducing Claude Opus 4.7Terminus-2 harness with thinking disabled, averaged over five attempts per task (footnote)Source image | Not yet |
| GPT-5.45 Mar 2026 · OpenAI | 75.1% | Introducing GPT-5.4harness not stated in the post | Not yet |
| Gemini 3.1 Pro19 Feb 2026 · Google | 68.5% | Gemini 3.1 Pro: A smarter model for your most complex tasksTerminus-2 harness; Thinking (High)Source image | Not yet |
| Claude Opus 4.65 Feb 2026 · Anthropic | 65.4% | Claude Opus 4.6Terminus-2 harness, 5-15 samples per task (footnote)Source image | Not yet |
| GPT-5.211 Dec 2025 · OpenAI | 62.2% | Introducing GPT-5.4reported later; harness not stated in the post | Not yet |
| Claude Opus 4.524 Nov 2025 · Anthropic | 59.3% | Introducing Claude Opus 4.5128K thinking budget, default effort (high), averaged over 5 trials (methodology)Source image | Not yet |
| Gemini 3 Pro18 Nov 2025 · Google | 54.2% | Start building with Gemini 3Terminus-2 agentSource image | Not yet |
| Gemini 3 Pro18 Nov 2025 · Google | 56.9% | Gemini 3.1 Pro: A smarter model for your most complex tasksreported later; another report, not charted; Terminus-2 harnessSource image | Not yet |
| Claude Sonnet 4.529 Sep 2025 · Anthropic | 50.0% | Introducing Claude Opus 4.5reported laterSource image | Not yet |
Benchmark
tbench.aiHigher is better