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
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 85.84 |
| 2 | Gemini 2.5 Flash | 83.66 |
| 3 | GPT-4o Audio | 81.28 |
| 4 | Gemini 3 Flash | 77.25 |
| 5 | GPT Audio | 71.28 |
| 6 | MERaLiON-2-10B | 62.94 |
| 7 | Qwen2-Audio-7B-Instruct | 42.09 |
| 8 | Gemma 3n E4B (IT) | 28.3 |
| 9 | SALMONN-13B | 28.11 |
Interactive version: theaggregate.ai/benchmark?slug=rail-memory · How It Works · Data refreshed daily, snapshot 2026-09-29.