Metric-Bench - 3D Bounding Box: leaderboard

Metric: Mean relative accuracy (%; 3D bounding-box questions; 1,340 template questions over ScanNetV2 indoor test scenes, each giving the real metric values of reference objects and asking for a target object's metric in camera coordinates; zero-shot; mean relative accuracy over thresholds 0.50-0.95; GPT-5-mini and Gemini-2.5-flash on a random 400-question subset). Source: arxiv.org. Saturation forecast: Around April 2028. 15 models tracked.

Top models

#ModelScore
1GPT-5 Mini33.4
2Qwen 3 VL 8B29.73
3Gemini 2.5 Flash29.09
4Qwen 2.5 VL 32B Instruct28.07
5Qwen 2.5 VL 7B Instruct19.84

Interactive version: theaggregate.ai/benchmark?slug=metric-bench-3d-bounding-box · How It Works · Data refreshed daily, snapshot 2026-09-26.