Metric-Bench - Depth: leaderboard

Metric: Mean relative accuracy (%; object-depth 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 December 2026. 15 models tracked.

Top models

#ModelScore
1Gemini 2.5 Flash61.64
2GPT-5 Mini58.88
3Qwen 2.5 VL 32B Instruct50.98
4Qwen 2.5 VL 7B Instruct47.34
5Qwen 3 VL 8B44.91

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