RAIL - Acquired Knowledge: leaderboard
Metric: LLM-as-judge score (%) on acquired-knowledge tasks over speech, sound and music; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 79.19 |
| 2 | Gemini 3 Flash | 76.63 |
| 3 | Gemini 2.5 Flash | 72.68 |
| 4 | GPT-4o Audio | 72.31 |
| 5 | MERaLiON-2-10B | 62.62 |
| 6 | Qwen2-Audio-7B-Instruct | 57.99 |
| 7 | GPT Audio | 51.53 |
| 8 | Gemma 3n E4B (IT) | 48.42 |
| 9 | SALMONN-13B | 32.35 |
Interactive version: theaggregate.ai/benchmark?slug=rail-acquired-knowledge · How It Works · Data refreshed daily, snapshot 2026-09-29.