MemLens (128K): leaderboard

MemLens tests the multimodal long-term memory of vision-language models with 789 questions whose answers depend on images and facts spread over many sessions. Overall accuracy in percent at a 128K-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: Overall accuracy (%) on all 789 questions (information extraction, multi-session, temporal and knowledge-update reasoning and answer refusal) at a 128K-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 February 2027. 25 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)51.99#99
2Kimi K2.551.99#266
3GPT-5.449.56#135
4Qwen 3 VL 235B A22B Instruct47.66#532
5Qwen 3.5 122B A10B45.5#346
6Qwen 3 VL 30B A3B Instruct44.23#895
7Qwen 3.5 27B40.68#375
8GLM-4.6V39.04#710
9Qwen 3 VL 8B Instruct34.73#1029
10Qwen 3 VL 235B A22B (Thinking)34.73#459
11Qwen 3.5 9B32.57#828
12Cosmos-Reason2-8B31.94#755
13Qwen 3 VL 30B A3B (Thinking)31.31#719
14Qwen 3 VL 8B (Thinking)30.29#910
15Qwen 3 VL 4B Instruct28.14#1332

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

Interactive version: theaggregate.ai/benchmark?slug=memlens-128k · How It Works · Data refreshed daily, snapshot 2026-11-21.