AstroMind - Threat Assessment: leaderboard
AstroMind (Liu et al., 2026) threat assessment: the model rates the threat that an inferred spacecraft maneuver poses to another nation's satellite (low, medium or high) with technical and strategic reasoning, judged against the reference by DeepSeek-R1. Accuracy in percent over 92 open-ended questions with a 4,096-token context; empty responses are excluded from scoring. Higher is better.
Metric: Accuracy (%; 92 open-ended threat-level assessments checked by a DeepSeek-R1 judge against the reference; empty responses excluded). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | QwQ-32B | 66.67 | #909 |
| 2 | GPT-OSS-20B | 59.09 | #1150 |
| 3 | Qwen 3 32B | 49.38 | #960 |
| 4 | Gemma 3 27B | 46.74 | #1638 |
| 5 | Qwen 3 8B | 45.45 | #1838 |
| 6 | Gemma 3 4B | 26.09 | #5104 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=astromind-threat-assessment · How It Works · Data refreshed daily, snapshot 2026-11-12.