ASTRA-bench: leaderboard
Metric: Task-completion score of a gpt-5.1-2025-11-13 judge against the human-authored success conditions (rescaled 0-1, shown times 100), macro-averaged over the low, moderate and high tranches of the referential, functional and informational complexity axes; 2,413 human-authored personal-assistant scenarios over the protagonists' email, calendar, messaging and phone histories in a stateful sandbox with 25+ tools; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 91.12 | #79 |
| 2 | DeepSeek V3.2 | 90.5 | #198 |
| 3 | O3 | 87.82 | #121 |
| 4 | GPT-4.1 | 79.22 | #240 |
| 5 | Claude Haiku 4.5 | 76.46 | #271 |
| 6 | Qwen 3 235B A22B | 69.02 | #304 |
| 7 | Qwen 3 30B A3B | 65.67 | #488 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=astra-bench · How It Works · Data refreshed daily, snapshot 2026-10-11.