MemLens (32K): leaderboard
Metric: Overall accuracy (%) on all 789 questions (information extraction, multi-session, temporal and knowledge-update reasoning and answer refusal) at a 32K-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 2029. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 122B A10B | 58.68 |
| 2 | Qwen 3 VL 30B A3B Instruct | 55.01 |
| 3 | Kimi K2.5 | 54.88 |
| 4 | Qwen 3 VL 235B A22B Instruct | 53.23 |
| 5 | Qwen 3.5 27B | 52.98 |
| 6 | GPT-5.4 | 52.72 |
| 7 | GLM-4.6V | 52.34 |
| 8 | Qwen 3.5 9B | 50.32 |
| 9 | Qwen 3 VL 8B Instruct | 49.18 |
| 10 | Qwen 3 VL 235B A22B (Thinking) | 48.29 |
| 11 | Qwen 3 VL 8B (Thinking) | 46.01 |
| 12 | Nemotron Nano 12B v2 VL | 44.99 |
| 13 | Qwen 3.5 4B | 44.36 |
| 14 | Qwen 3.5 2B | 44.36 |
| 15 | Qwen 3 VL 4B Instruct | 44.23 |
Interactive version: theaggregate.ai/benchmark?slug=memlens-32k · How It Works · Data refreshed daily, snapshot 2026-10-07.