RAIL - Auditory Processing: leaderboard

Metric: LLM-as-judge score (%) on auditory-processing tasks (speech-sound discrimination, distortion resistance, sound localization); GPT-5.4 judge; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 26 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)62.14
2Gemini 2.5 Flash53.76
3Gemini 3 Flash52.82
4GPT-4o Audio51.11
5MERaLiON-2-10B46.32
6Qwen2-Audio-7B-Instruct42.56
7Gemma 3n E4B (IT)36.75
8SALMONN-13B34.79
9GPT Audio26.32

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