DriveSpatial - Understanding: 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 multiple-choice spatial-understanding tasks on the full 15,670-question set, scored as accuracy in percent; higher is better.

Metric: Accuracy (%) on the multiple-choice spatial-understanding tasks (relations across viewpoints) 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; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 8B50.62#927
2InternVL3.5-8B44.15#1618
3Gemma 3 12B (IT)44.05#1907
4DriveSpatial SenseNova-SI (checkpoint unspecified)36.25
5LongVA-7B35.9#3480
6SpaceThinker31.74#1467
7DriveSpatial Ego3D (checkpoint unspecified)30.67
8LLaVA-OneVision-7B27.64#3002
9LLaVA-Video-7B27.57#2303
10InternVL3-8B27.26#1795
11DeepSeek-VL2-Small26.6#2523
12RoboTron-Drive21.03

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 · How It Works · Data refreshed daily, snapshot 2026-11-22.