Tau3-Voice Multilingual: leaderboard
Sierra's τ³-Voice leaderboard (taubench.com), Multilingual edition (τ³-Voice 2.0, the site's default view): voice agents handle τ³-bench customer-service calls with a simulated customer in six languages. Standard submissions only; incomplete submission packages are not shown. Score: task-completion pass@1 (%) over the airline, retail and telecom domains, with every language-domain cell weighted equally. Separate from the English edition. Higher is better.
Metric: Task completion pass@1 (%), six languages x airline/retail/telecom cells weighted equally. Source: taubench.com. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Ava Zero | 86 | |
| 2 | Phonic 1 (High) | 83.22 | |
| 3 | gpt-live-1 | 79.91 | |
| 4 | Grok Voice 2.0 (High) | 71.37 | |
| 5 | Gemini 3.8 Live (Extended Thinking) | 68.51 |
Interactive version: theaggregate.ai/benchmark?slug=tau3-voice-multilingual · How It Works · Data refreshed daily, snapshot 2026-11-05.