M3Exam (Multimodal Memory) - Multimodal Reasoning: leaderboard

Metric: LLM-judge score (0-100) on the chart-grounded multimodal reasoning (MR) questions of 5,150 questions over 239 long multi-session user-assistant conversations with images and PDFs (15 personas); each model answers with the annotated gold supporting sessions and their images or rendered pages placed in context, no memory module; scored by a Qwen2.5-VL-32B-Instruct judge on a five-level rubric (0, 0.25, 0.5, 0.75, 1), printed 0-1 and x100 here; higher is better. Source: arxiv.org. Saturation forecast: Around October 2027. 9 models tracked.

Top models

#ModelScore
1Claude Opus 4.673.06
2Grok 470.34
3Seed 2.0 Pro65.86
4Kimi K2.565.56
5GLM-5.165.5
6Qwen 3.6 Plus64.06
7GPT-5.463.89
8Gemini 3.1 Pro (Preview)51.11
9Qwen 2.5 VL 7B Instruct46.82

Interactive version: theaggregate.ai/benchmark?slug=m3exam-multimodal-memory-multimodal-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-29.