MemLens (32K) - Temporal Reasoning: leaderboard

Accuracy in percent on the MemLens 194 questions about dates, durations and the order of events, at a 32K-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 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: Not forecast. 27 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 30B A3B Instruct60.82#895
2Qwen 3 VL 8B Instruct58.25#1029
3Qwen 3 VL 235B A22B Instruct55.67#532
4Kimi K2.553.09#266
5GLM-4.6V53.09#710
6Qwen 3 VL 8B (Thinking)52.58#910
7Qwen 3.5 122B A10B51.55#346
8Qwen 3 VL 2B Instruct50.52#2367
9Qwen 3 VL 235B A22B (Thinking)49.48#459
10Cosmos-Reason2-8B48.45#755
11Nemotron Nano 12B v2 VL44.85#2368
12Gemma 3 12B (IT)43.81#1887
13Phi-4 Multimodal Instruct43.81#4154
14Qwen 3 VL 4B (Thinking)43.3#1334
15Qwen 3 VL 4B Instruct42.78#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-32k-temporal-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.