M3Exam (Multimodal Memory) - Thematic Reasoning: leaderboard
M3Exam (2026) is a multimodal conversational-memory benchmark of 5,150 questions over long multi-session user-assistant conversations with images and documents; the frontier models here answer with the gold supporting sessions in context. This board is the LLM-judge score (0-100) on the thematic reasoning questions; higher is better.
Metric: LLM-judge score (0-100) on the thematic reasoning (TH) 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 February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.1 | 61.93 | #179 |
| 2 | Claude Opus 4.6 | 60.83 | #102 |
| 3 | Qwen 3.6 Plus | 52.27 | #222 |
| 4 | Gemini 3.1 Pro (Preview) | 44.17 | #94 |
| 5 | Kimi K2.5 | 43.51 | #256 |
| 6 | GPT-5.4 | 43.33 | #123 |
| 7 | Grok 4 | 36.67 | #307 |
| 8 | Seed 2.0 Pro | 13.64 | #211 |
| 9 | Qwen 2.5 VL 7B Instruct | 6.35 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=m3exam-multimodal-memory-thematic-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-13.