VEHBench - Corrupted-State Recovery: leaderboard
Metric: P3 success (%; share of corrupted or misleading design trajectories that end feasible within the recovery budget; LLM-assisted vibration energy harvester design against a physics oracle; one complete run per model). Source: arxiv.org. Saturation forecast: Around January 2028. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Hy3-preview (Reasoning) | 47.4 |
| 2 | MiMo-V2.5-Pro (Non-reasoning) | 45.5 |
| 3 | DeepSeek R1 | 42.9 |
| 4 | GPT-5.4 | 42.3 |
| 5 | Gemini 3.1 Pro (Preview) | 37.2 |
| 6 | DeepSeek V3 | 35.3 |
| 7 | Qwen 3 Max | 30.1 |
| 8 | DeepSeek V4 Pro (Non-reasoning) | 28.8 |
| 9 | Qwen 3.6 Plus | 27.6 |
| 10 | O4 Mini | 26.3 |
| 11 | Claude Sonnet 4.6 | 16 |
| 12 | Llama 3.3 70B | 3.2 |
Interactive version: theaggregate.ai/benchmark?slug=vehbench-corrupted-state-recovery · How It Works · Data refreshed daily, snapshot 2026-09-29.