AeroCopilotBench - Tier-2 Safety-Gated Outcome: leaderboard
AeroCopilotBench (arXiv 2608.16349v1): LLM agents executing Cessna 172S and Piper PA-44 emergency and abnormal procedures in an interactive virtual cockpit, 73 Tier-2 tasks with three episodes each, thinking enabled at high effort. Mean goal attainment with primary-goal gating, set to zero for episodes with any safety violation; percent, higher is better.
Metric: Goal attainment (%; zero on any safety violation). Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (High) | 91.1 | #22 (GPT-5.6 Sol) |
| 2 | Gemini 3.5 Flash (High) | 84.4 | #66 (Gemini 3.5 Flash) |
| 3 | GLM-5.1 (Thinking) | 83.1 | #109 (GLM-5.1) |
| 4 | Qwen 3.7 Max (Thinking) | 80.5 | #85 (Qwen 3.7 Max) |
| 5 | GLM-5.2 (High) | 78.4 | #80 (GLM-5.2) |
| 6 | DeepSeek V4 Pro (High) | 76.6 | #112 (DeepSeek V4 Pro) |
| 7 | DeepSeek V4 Flash (High) | 67.7 | #155 (DeepSeek V4 Flash) |
| 8 | Kimi K2.6 (Thinking) | 67.5 | #113 (Kimi K2.6) |
| 9 | Qwen 3.5 397B A17B (Thinking) | 54.7 | #154 (Qwen 3.5 397B A17B) |
| 10 | MiniMax-M2.5 | 47.5 | #316 |
| 11 | Nex N2 Pro | 45.9 | #117 |
| 12 | Qwen 3.5 122B A10B (Thinking) | 45.8 | #184 (Qwen 3.5 122B A10B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aerocopilotbench-tier-2-safety-gated-outcome · How It Works · Data refreshed daily, snapshot 2026-11-05.