AA Speech to Speech - Tau3-Voice: leaderboard
Artificial Analysis speech-to-speech page: AA's own runs of Sierra's τ³-Voice customer-service tasks (retail, airline, telecom) with live voice models; score is the reported overall pass^1 across retail, airline and telecom. Separate from Sierra's own τ³-Voice leaderboard boards, which list submitted runs. One component of AA's Speech to Speech Index. Pass^1 (%), higher is better.
Metric: Reported overall pass^1 (%), AA's own runs across retail, airline and telecom. Source: artificialanalysis.ai. Saturation forecast: Rough model projection: around 2026. 29 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Phonic 1 (High) | 79.93 | |
| 2 | GPT-Live-1 (Astra, medium) | 74.48 | |
| 3 | Grok Voice 3.0 | 71.65 | |
| 4 | Gemini 3.8 Live Extended Thinking (High) | 68.64 | |
| 5 | GPT-Live-1 (Sol, low) | 59.25 | |
| 6 | Grok Voice 2.0 (High) | 56.48 | |
| 7 | Qwen Audio 3.0 Realtime Plus | 54.6 | |
| 8 | Grok Voice 1.0 | 52.08 | |
| 9 | Eleven Realtime v1 Max | 47.61 | |
| 10 | Eleven Realtime v1 Mini | 46.79 |
Interactive version: theaggregate.ai/benchmark?slug=aa-speech-to-speech-tau3-voice · How It Works · Data refreshed daily, snapshot 2026-11-05.