AudioRAG: leaderboard
Metric: Accuracy (%) on AudioRAG's 500 multi-hop questions about a sound, speech or music clip that also need facts beyond the clip (LLM-generated from dataset metadata or manually curated from web audio, then filtered by search agents and annotators); raw model answering from the audio and its own knowledge, no retrieval; a GPT-4o judge compares each answer with the reference; mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 45 | #237 |
| 2 | Qwen2.5-Omni-7B | 32.2 | #601 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=audiorag · How It Works · Data refreshed daily, snapshot 2026-10-11.