FLY-EVAL++ - History-Conditioned Prediction - Total: leaderboard
FLY-EVAL++ (arXiv 2609.04021v1, Table 1): history-conditioned one-step flight-state prediction (M1) total score from deterministic protocol, field, physics, safety and accuracy verifiers. Percent; higher is better. Near-saturated in the paper (standard deviation 0.53 across models). Only the 21 of 66 screened models above 90% protocol compliance are listed.
Metric: M1 total score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 97.31 | #105 |
| 2 | Gemini 3 Pro | 97.23 | #82 |
| 3 | DeepSeek V3 | 97.14 | #329 |
| 4 | DeepSeek V3.2 Exp | 97.08 | #241 |
| 5 | Qwen 2.5 32B | 97.02 | #544 |
| 6 | Qwen 3 235B A22B | 97.01 | #323 |
| 7 | Qwen 3 32B | 96.98 | #449 |
| 8 | Claude 3.7 Sonnet (20250219) | 96.96 | #209 |
| 9 | Qwen3-Next-80B | 96.94 | #321 |
| 10 | GPT-4o | 96.92 | #357 |
| 11 | Llama 3.1 405B | 96.85 | #476 |
| 12 | Claude Sonnet 4.5 | 96.84 | #148 |
| 13 | O4 Mini | 96.79 | #185 |
| 14 | Kimi K2 (Thinking) | 96.77 | #252 (Kimi K2) |
| 15 | Gemini 2.5 Pro | 96.76 | #159 |
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-history-conditioned-prediction-total · How It Works · Data refreshed daily, snapshot 2026-11-05.