M3Exam (Multimodal Memory) - Implicit Inference: leaderboard

Metric: LLM-judge score (0-100) on the implicit inference (II) 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 December 2026. 9 models tracked.

Top models

#ModelScore
1GPT-5.471.25
2Gemini 3.1 Pro (Preview)68.75
3Grok 468.75
4Claude Opus 4.665.01
5GLM-5.156.67
6Kimi K2.555.89
7Seed 2.0 Pro41.03
8Qwen 3.6 Plus38.33
9Qwen 2.5 VL 7B Instruct5.23

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