ATC Safety-Oriented Language Understanding Eval: leaderboard
Safety-oriented air-traffic-control language-understanding evaluation for structured ATC utterances, consequence-aware entity/action scoring, and operational-risk sensitivity.
Metric: Risk Score (imported). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Around January 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 69.22 | #82 |
| 2 | Gemini 3 Pro (Preview) | 68.84 | #66 |
| 3 | GPT-5.1 | 67.75 | #135 |
| 4 | Qwen 3 Max | 52.34 | #208 |
| 5 | Grok 4.1 Fast (Reasoning) | 52.06 | #217 (Grok 4.1 Fast) |
| 6 | GPT-4o Mini | 44.83 | #599 |
| 7 | Qwen Plus | 43.52 | #385 |
| 8 | Qwen 3 32B | 42.49 | #443 |
| 9 | DeepSeek V3 Chat | 40.03 | #447 |
| 10 | Qwen 3 8B | 28.8 | #674 |
| 11 | Qwen 3 14B | 25.85 | #537 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=atc-safety-oriented-language-understanding-eval · How It Works · Data refreshed daily, snapshot 2026-10-20.