RAIL - Memory: leaderboard

Metric: LLM-as-judge score (%) on auditory memory tasks (memory span, associative memory, sound-pattern memory); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 26 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)85.84
2Gemini 2.5 Flash83.66
3GPT-4o Audio81.28
4Gemini 3 Flash77.25
5GPT Audio71.28
6MERaLiON-2-10B62.94
7Qwen2-Audio-7B-Instruct42.09
8Gemma 3n E4B (IT)28.3
9SALMONN-13B28.11

Interactive version: theaggregate.ai/benchmark?slug=rail-memory · How It Works · Data refreshed daily, snapshot 2026-09-29.