MM-JudgeBias - Image Dominance: leaderboard

MM-JudgeBias (2026) tests multimodal LLM judges for compositional bias: the judge scores image, query and response triplets from 1 to 10 before and after a controlled perturbation. Bias-Deviation from 0 to 1, the normalized score drop when the query is emptied; higher is better.

Metric: Bias-Deviation (0 to 1) on MM-JudgeBias (about 200 image-query-response triplets per bias type, 1,804 in all, each judged on a 1-10 scale before and after a controlled perturbation; mean of three runs): mean drop of the judge's score, normalized by the largest possible drop, when the query is replaced with empty text; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 26 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.5Thinking0.589#165
2Claude Sonnet 4.5Thinking0.54#267
3Gemini 2.5 Pro0.535#280
4Claude Haiku 4.5Thinking0.53#587
5O4 MiniHigh0.446#372
6Gemini 2.5 Flash Lite0.367#972
7Gemini 2.0 Flash Lite0.358#1064
8O3High0.354#242
9Qwen 3 VL 30B A3B (Thinking)0.343#727
10Prometheus-Vision-13B0.34
11Gemini 2.5 Flash0.292#466
12Gemini 3 ProHigh0.278#146
13Qwen 3 VL 8B Instruct0.266#1043
14LLaVA-Critic-7B0.266
15InternVL3.5-14B0.243#1519

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mm-judgebias-image-dominance · How It Works · Data refreshed daily, snapshot 2026-11-22.