MemLens (128K) - Information Extraction: leaderboard
Metric: Accuracy (%) on the information extraction (IE, 246 questions) 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 January 2027. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 60.16 |
| 2 | Qwen 3 VL 30B A3B Instruct | 56.5 |
| 3 | Gemini 3.1 Pro (Preview) | 55.79 |
| 4 | Qwen 3.5 27B | 54.07 |
| 5 | GLM-4.6V | 52.85 |
| 6 | Kimi K2.5 | 51.63 |
| 7 | Qwen 3 VL 235B A22B Instruct | 51.22 |
| 8 | Qwen 3.5 9B | 45.12 |
| 9 | Qwen 3.5 122B A10B | 43.09 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 33.74 |
| 11 | Qwen 3 VL 8B (Thinking) | 33.33 |
| 12 | Qwen 3 VL 8B Instruct | 32.52 |
| 13 | Qwen 3.5 4B | 29.67 |
| 14 | Qwen 3 VL 235B A22B (Thinking) | 29.27 |
| 15 | Qwen 3 VL 4B Instruct | 28.46 |
Interactive version: theaggregate.ai/benchmark?slug=memlens-128k-information-extraction · How It Works · Data refreshed daily, snapshot 2026-10-07.