DriveSpatial - 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 the full 15,670-question set, 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 the full benchmark (15,670 human-verified questions over 20 tasks from nuScenes, Waymo, TruckScenes, AV2 and ONCE); open-source generalist and driving or spatial specialist VLMs; lower is better. Source: arxiv.org. Saturation forecast: Around May 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1InternVL3.5-8B14.26#1618
2Qwen 3 VL 8B14.32#927
3Gemma 3 12B (IT)14.65#1907
4DeepSeek-VL2-Small15.15#2523
5DriveSpatial SenseNova-SI (checkpoint unspecified)15.25
6RoboTron-Drive15.49
7LLaVA-Video-7B15.68#2303
8LLaVA-OneVision-7B15.76#3002
9InternVL3-8B15.95#1795
10DriveSpatial Ego3D (checkpoint unspecified)15.95
11SpaceThinker16.96#1467
12LongVA-7B20.26#3480

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=drivespatial-understanding-rmse · How It Works · Data refreshed daily, snapshot 2026-11-22.