MemLens (128K) - Temporal Reasoning: leaderboard

Accuracy in percent on the MemLens 194 questions about dates, durations and the order of events, 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: Accuracy (%) on the temporal reasoning (TR, 194 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 August 2027. 24 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 235B A22B Instruct54.64#532
2Qwen 3 VL 30B A3B Instruct52.58#895
3Qwen 3 VL 8B Instruct44.85#1029
4GLM-4.6V43.81#710
5Cosmos-Reason2-8B43.81#755
6Qwen 3 VL 235B A22B (Thinking)43.3#459
7Qwen 3.5 122B A10B41.75#346
8Gemini 3.1 Pro (Preview)41.24#99
9GPT-5.439.69#135
10Qwen 3 VL 8B (Thinking)39.18#910
11Gemma 3 27B (IT)37.11#1324
12Qwen 3 VL 30B A3B (Thinking)36.6#719
13Qwen 3 VL 4B Instruct35.05#1332
14Qwen 3.5 27B34.02#375
15Claude Sonnet 4.532.99#261

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-temporal-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.