DriveSpatial-mini - Understanding RMSE: leaderboard

DriveSpatial (2026) tests whether vision-language models can build and reason over a coherent driving scene from synchronized multi-view camera frames, with human-verified questions generated from scene graphs of nuScenes, Waymo, TruckScenes, AV2 and ONCE. This board covers the numeric spatial-understanding tasks on DriveSpatial-mini, the stratified 1,000-question subset the paper uses for proprietary models and humans, scored as root mean squared error of numeric answers such as distances and counts; lower is better.

Metric: Root mean squared error of numeric answers (distances, counts and other spatial measurements) on the spatial-understanding tasks on DriveSpatial-mini (a stratified 1,000-question subset, 50 per task), where the paper evaluates the proprietary VLMs and humans; lower is better. Source: arxiv.org. Saturation forecast: Around March 2028. 3 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-510.41#178
2GPT-4o12.71#751
3DriveSpatial Gemini-2 Pro (checkpoint unspecified)13.2

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

Interactive version: theaggregate.ai/benchmark?slug=drivespatial-mini-understanding-rmse · How It Works · Data refreshed daily, snapshot 2026-11-21.