Pricing
Only pay for the audio you send us and the audio we generate.
Chat with us if you need an organization account for Enterprise use.
What you would pay
Competitor figures are public list prices for the closest comparable product, last checked June to September 2026. Where a provider sells capacity by subscription rather than by the hour, we use their best published tier. Committed use and enterprise agreements price differently on both sides. Speech to text compares against batch rates, which is the cheaper end of most of this field.
You save
less than OpenAI at 1,000 hours of audio.
Rates
Batch rates. A realtime session over a WebSocket bills at 25% over these rates on Rhythm and Symphony, except speech out, which costs the same either way.
Audio is one token per 100 ms, so 36,000 tokens an hour, and an hour of transcribed speech returns roughly 15,000 output text tokens. An hour of synthesized speech consumes roughly 60,000 characters, on our models and on every provider priced per character. Token rates are only comparable between providers counting tokens the same way, which is why the hour leads.
See all model detailsAccounts and limits
Everything that is not a per-request rate. If your answer is not here, talk to us.