MemLens (64K) - Knowledge Update: leaderboard

Accuracy in percent on the MemLens 116 questions whose answer changed in a later session, at a 64K-token interleaved image and text conversation history, with the evidence sessions hidden among distractor sessions (MemLens, 2026), judged by Qwen3-VL-235B. Higher is better.

Metric: Accuracy (%) on the knowledge update (KU, 116 questions) 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 March 2027. 25 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Kimi K2.549.14#266
2Qwen 3.5 27B43.97#375
3GPT-5.443.1#135
4Qwen 3 VL 30B A3B Instruct39.66#895
5Qwen 3.5 9B37.93#828
6Qwen 3 VL 235B A22B Instruct36.21#532
7Qwen 3 VL 235B A22B (Thinking)35.34#459
8GLM-4.6V34.48#710
9Nemotron Nano 12B v2 VL34.48#2368
10Qwen 3 VL 8B (Thinking)30.17#910
11Qwen 3 VL 8B Instruct29.31#1029
12Qwen 3 VL 30B A3B (Thinking)29.31#719
13Qwen 3.5 4B28.45#1241
14Gemma 3 27B (IT)27.59#1324
15Phi-4 Multimodal Instruct24.14#4154

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=memlens-64k-knowledge-update · How It Works · Data refreshed daily, snapshot 2026-11-21.