PilotBench - Regression Score: leaderboard

PilotBench (2026) evaluates LLMs as general-aviation flight predictors: from structured historical waypoints of real trajectories across nine flight phases, a model predicts the next trajectory and attitude state in a fixed output schema at temperature 0. This board is the regression score on a 0 to 100 scale, a piecewise-linear transform of the prediction's mean absolute and root mean square errors; higher is better.

Metric: Regression score (0-100): 0.6 x piecewise-linear MAE score plus 0.4 x RMSE score of the predicted flight state, on PilotBench (one-step trajectory and attitude prediction from 34-channel telemetry of 708 general-aviation trajectories in nine flight phases, fixed output schema, temperature 0); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1QwQ-32B85.53#907
2Qwen 3 32B85.43#972
3O3 Mini84.24#546
4Doubao-1.5-Pro-32k83.53#694
5Qwen 2.5 72B Instruct83.2#1117
6DeepSeek V383.19#729
7GPT-4o Mini82.99#1667
8Qwen 2.5 32B Instruct79.01#1247
9Qwen 2.5 14B Instruct76.31#1775
10Qwen 2.5 VL 72B Instruct65.66#841
11Qwen 2.5 7B Instruct54.98#2723
12DeepSeek R1 Distill Llama 70B52.77#1669
13glm-4-9B50.75#2738
14DeepSeek R1 Distill Qwen 32B38.24#1834
15GLM-Z1-Rumination-32B32.23

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=pilotbench-regression-score · How It Works · Data refreshed daily, snapshot 2026-11-21.