FLY-EVAL++ - Single-Step Prediction - Protocol Compliance: leaderboard
FLY-EVAL++ (arXiv 2609.04021v1, Table 1) single-step flight-state prediction (S1), D1 protocol score: valid JSON with all required fields. 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 D1 protocol score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet (20250219) | 94.8 | #209 |
| 2 | Claude Sonnet 4.5 | 94.5 | #148 |
| 3 | DeepSeek R1 0528 | 94.4 | #232 |
| 4 | O4 Mini | 94.1 | #185 |
| 5 | Qwen 3 32B | 93.4 | #449 |
| 6 | Qwen 3 235B A22B | 93.4 | #323 |
| 7 | Qwen 2.5 32B | 93.4 | #544 |
| 8 | Kimi K2 0905 | 93.1 | #292 |
| 9 | Qwen3-Next-80B | 93.1 | #321 |
| 10 | Llama 3.1 405B | 92.9 | #476 |
| 11 | DeepSeek V3.2 Exp | 92.5 | #241 |
| 12 | Gemini 3 Pro | 92.4 | #82 |
| 13 | Kimi K2 (Thinking) | 92.4 | #252 (Kimi K2) |
| 14 | DeepSeek V3.1 | 92.1 | #275 |
| 15 | GPT-5 | 91.9 | #105 |
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-protocol-compliance · How It Works · Data refreshed daily, snapshot 2026-11-05.