Terminal-Bench Science 0.1
- OpenAI
- Anthropic
Results
| Model | Score | Source | Human-verified |
|---|---|---|---|
| Gemini 4 Argon30 Sep 2026 · Google | 57.6% | Gemini 4 Argon: our next era of frontier intelligenceSource image | Yes |
| Claude Opus 5.522 Sep 2026 · Anthropic | 58.7% | Introducing Claude Opus 5.5max effort; standard error ±3.5–5 pts | Yes |
| GPT-6 Astra3 Sep 2026 · OpenAI | 64.6% | GPT-6 Astra: A new generation of intelligence | Yes |
Benchmark
tbench.aiHigher is better