AstaBench - Code & Execution: leaderboard
Agents set up and run research repositories (SUPER-Expert), reproduce published computational results (CORE-Bench-Hard) and complete data-science code (DS-1000); the score is the mean of the three. Unlike HAL's CORE-Bench Hard board it averages reproduction with repository setup and library-coding tasks under each agent's own scaffold.
Metric: Mean of SUPER-Expert, CORE-Bench-Hard and DS-1000 test scores (0-1). Source: allenai-asta-bench-leaderboard.hf.space. Saturation forecast: Around February 2027. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (xHigh) | 0.7 | #79 (Claude Opus 4.7) |
| 2 | GPT-5.4 (2026-03-05) (xHigh) | 0.69 | #87 (GPT-5.4 (2026-03-05)) |
| 3 | GPT-5 | 0.6 | #167 |
| 4 | GPT-5 (Medium) | 0.55 | #167 (GPT-5) |
| 5 | GPT-5 Mini (2025-08-07) (Medium) | 0.51 | #343 (GPT-5 Mini (2025-08-07)) |
| 6 | O3 (2025-04-16) (Medium) | 0.49 | #216 (O3 (2025-04-16)) |
| 7 | Claude Sonnet 4 (20250514) (Thinking) | 0.46 | #387 (Claude Sonnet 4 (20250514)) |
| 8 | GPT-4.1 | 0.32 | #495 |
| 9 | Gemini 2.5 Flash (Preview 05-20) | 0.26 | #426 |
| 10 | Claude 3.5 Haiku (20241022) | 0.22 | #1578 |
| 11 | GPT-4o (2024-08-06) | 0.18 | #740 |
| 12 | Llama 4 Scout Instruct | 0.05 | #1543 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=astabench-code-execution · How It Works · Data refreshed daily, snapshot 2026-11-13.