SUPER-Expert (AstaBench 2026)
Code · 2026-04-30
Research-code execution on the SUPER-Expert AstaBench test split. The component reports partial progress as a mean task score rather than a binary full-task success rate.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.7 | 0.4 |
| Sonnet 4.6 | 0.4 |
| GPT-5 | 0.4 |
| GPT-5.5 | 0.4 |
| Opus 4.6 | 0.4 |
| GPT-5 Mini | 0.3 |
| Gemini 3.1 Pro Preview | 0.3 |
| GPT-5.4 | 0.3 |
| Sonnet 4 | 0.2 |
| o3 | 0.2 |
| Haiku 3.5 | 0.1 |
| GPT-4.1 | 0.1 |
| GPT-4o (2024-08-06) | 0.1 |