Metric-Bench - Size: leaderboard

Metric: Mean relative accuracy (%; length, width, height, elevation and dimension 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 July 2028. 15 models tracked.

Top models

#ModelScore
1Qwen 3 VL 8B40.44
2GPT-5 Mini38.47
3Gemini 2.5 Flash34.74
4Qwen 2.5 VL 32B Instruct33.82
5Qwen 2.5 VL 7B Instruct27.75

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