MMBU - Grounded Classification (Boxes): leaderboard

Metric: Closed-ended classification of a region marked by a bounding box (from detection annotations): micro-averaged F1 x100 (0-100) over MMBU's biomedical images from 35 submodalities (95% bootstrap intervals not published); zero-shot, official checkpoints; exact match after normalization. Source: arxiv.org. Saturation forecast: Around July 2028. 17 models tracked.

Top models

#ModelScore
1InternVL3.5-8B43.9
2Qwen 3 VL 32B Instruct43.3
3GPT-4.1 Mini43
4GPT-5.4 Mini41
5Qwen 2.5 VL 32B Instruct40.1
6Qwen 3 VL 8B Instruct39.1
7MedGemma-4B-IT37.1
8Lingshu-32B36.8
9Qwen 3 VL 4B Instruct36.3
10Lingshu-7B36
11Qwen 2.5 VL 7B Instruct32
12Gemma 3 4B (IT)28.3

Interactive version: theaggregate.ai/benchmark?slug=mmbu-grounded-classification-boxes · How It Works · Data refreshed daily, snapshot 2026-09-29.