A round of voice upgrades has landed across the platform.

Realtime

Live voice conversations now run against speech-native models with proper interruption handling. Sessions are billed by audio time with the rate shown before you connect, and idle sessions wind down automatically — a forgotten tab cannot quietly keep billing you.

Text-to-speech

The TTS roster grew across providers, from fast utility voices to expressive narration voices. The voice picker shows the price next to each voice, so choosing a premium voice is an informed decision, not a surprise on the ledger.

Transcription

Speech-to-text now routes to fast inference providers when speed is what matters, with higher-fidelity models available when accuracy is critical. If a provider fails mid-request, the platform falls back instead of dropping your audio.

Billing

All voice usage flows through the same ledger as everything else: one billing record per call, priced at the moment of use.

Try it from the chat toolbar or the dedicated Live pages, and report anything odd in the Troubleshooting category.