VEHBench - Policy-Conditioned Selection: leaderboard
VEHBench (2026) is a stage-local diagnostic benchmark for LLM-assisted design of vibration energy harvesters for battery-free IoT, scored against a physics oracle. Kendall tau-b from -1 to 1 between the model ranking of feasible candidates and the oracle ranking under a stated deployment policy; higher is better.
Metric: P4 Kendall tau-b (-1 to 1): rank agreement between the model ranking of a feasible candidate pool and the oracle ranking under an explicit deployment policy; LLM-assisted vibration energy harvester design against a physics oracle; one complete run per model. Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 0.89 | #123 |
| 2 | Qwen 3.6 Plus | 0.88 | #222 |
| 3 | DeepSeek V3 | 0.86 | #716 |
| 4 | MiMo-V2.5-Pro (Non-reasoning) | 0.84 | #279 (MiMo-V2.5-Pro) |
| 5 | Claude Sonnet 4.6 | 0.84 | #161 |
| 6 | Hy3-preview (Reasoning) | 0.84 | #316 (Hy3-preview) |
| 7 | Qwen 3 Max | 0.84 | #403 |
| 8 | DeepSeek R1 | 0.83 | #473 |
| 9 | Gemini 3.1 Pro (Preview) | 0.82 | #94 |
| 10 | DeepSeek V4 Pro (Non-reasoning) | 0.79 | #188 (DeepSeek V4 Pro) |
| 11 | O4 Mini | 0.78 | #353 |
| 12 | Llama 3.3 70B | 0.71 | #1498 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vehbench-policy-conditioned-selection · How It Works · Data refreshed daily, snapshot 2026-11-13.