AA Speech to Speech - Big Bench Audio: leaderboard
Artificial Analysis speech-to-speech page: Big Bench Audio, spoken versions of Big Bench Hard reasoning questions answered by live voice models over audio, as run by AA. One component of AA's Speech to Speech Index. Accuracy (%), higher is better.
Metric: Accuracy (%). Source: artificialanalysis.ai. Saturation forecast: Rough model projection: around 2026. 44 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | StepAudio 3 Realtime (AA system stepfun_stepaudio-3-realtime) | 99.7 | |
| 2 | Qwen Audio 3.0 Realtime Plus | 99.2 | |
| 3 | Phonic 1 (High) | 98.9 | |
| 4 | Eleven Realtime v1 Max | 98.8 | |
| 5 | Qwen3.5 Omni Plus Realtime | 98.7 | |
| 6 | Grok Voice 3.0 | 98.5 | |
| 7 | Gemini 3.8 Live Extended Thinking (High) | 97.7 | |
| 8 | Step-Audio R1.1 (Realtime) | 97.6 | |
| 9 | Grok Voice 2.0 (High) | 97.2 | |
| 10 | Grok Voice 1.0 | 97.1 |
Interactive version: theaggregate.ai/benchmark?slug=aa-speech-to-speech-big-bench-audio · How It Works · Data refreshed daily, snapshot 2026-11-05.