FLY-EVAL++ - Single-Step Prediction - Physics Consistency: leaderboard
FLY-EVAL++ (arXiv 2609.04021v1, Table 1) single-step flight-state prediction (S1), D3 physics score: aerodynamic and cross-field consistency rules. Deterministic evidence-based verification of 45-field JSON predictions; percent, higher is better. Only the 21 of 66 screened models above 90% protocol compliance are listed.
Metric: S1 D3 physics score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 100 | #105 |
| 2 | DeepSeek V3 | 100 | #329 |
| 3 | Seed-1.6 | 100 | #272 |
| 4 | GPT-4o | 99.9 | #357 |
| 5 | Gemini 3 Pro | 99.9 | #82 |
| 6 | Gemini 2.5 Pro | 99.7 | #159 |
| 7 | Llama 3.1 405B | 99.7 | #476 |
| 8 | Grok 4 | 99.6 | #189 |
| 9 | DeepSeek R1 0528 | 99.4 | #232 |
| 10 | Kimi K2 (Thinking) | 99.3 | #252 (Kimi K2) |
| 11 | DeepSeek V3.1 | 98.9 | #275 |
| 12 | O4 Mini | 98.7 | #185 |
| 13 | Kimi K2 0905 | 98.6 | #292 |
| 14 | DeepSeek V3.2 Exp | 98.4 | #241 |
| 15 | Qwen3-Next-80B | 96.9 | #321 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=fly-eval-plus-plus-single-step-prediction-physics-consistency · How It Works · Data refreshed daily, snapshot 2026-11-05.