MemLens (64K): leaderboard
Metric: Overall accuracy (%) on all 789 questions (information extraction, multi-session, temporal and knowledge-update reasoning and answer refusal) at a 64K-token history, interleaved multimodal multi-session conversation history with the evidence sessions hidden among haystack sessions, judged correct or incorrect by a Qwen3-VL-235B-A22B-Instruct judge; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 122B A10B | 55.89 |
| 2 | Kimi K2.5 | 53.99 |
| 3 | Gemini 3.1 Pro (Preview) | 52.99 |
| 4 | GPT-5.4 | 52.34 |
| 5 | Qwen 3 VL 235B A22B Instruct | 50.82 |
| 6 | Qwen 3 VL 30B A3B Instruct | 49.81 |
| 7 | GLM-4.6V | 49.05 |
| 8 | Qwen 3.5 27B | 46.13 |
| 9 | Qwen 3.5 9B | 44.36 |
| 10 | Qwen 3 VL 8B Instruct | 42.84 |
| 11 | Nemotron Nano 12B v2 VL | 40.94 |
| 12 | Qwen 3 VL 8B (Thinking) | 39.67 |
| 13 | Qwen 3 VL 235B A22B (Thinking) | 39.67 |
| 14 | Qwen 3.5 4B | 37.14 |
| 15 | Qwen 3 VL 4B Instruct | 37.01 |
Interactive version: theaggregate.ai/benchmark?slug=memlens-64k · How It Works · Data refreshed daily, snapshot 2026-10-07.