AssemblyAI
4.5
AssemblyAI provides speech-to-text for pre-recorded and streaming audio, speech understanding add-ons, guardrails such as PII redaction, a text-to-speech API at $15 per million characters, and an LLM gateway. Pre-recorded Universal-2 is $0.15 an hour and Universal-3.5 Pro $0.21; streaming is billed on connection time at $0.15 to $0.45 an hour. New accounts get $50 in free credits.
- Commercial use OK
- Free tier
- Enterprise contract
- Pricing
- 従量課金
- Free tier
- Free tier
- Commercial use
- Permitted
- Japanese UI
- Not confirmed
- Enterprise contract
- Available
What it is good for
- Transcription that also needs summarisation or topic detection
- Pipelines requiring PII redaction before storage
- Per-hour pricing that is easier to reason about than per-minute
- Teams wanting one key for both transcription and LLM calls
Strengths and trade-offs
Strengths
- PII redaction built in rather than a downstream step
- Per-hour pricing is simpler to estimate than per-minute
- $50 free credit with no card required
- Speech understanding add-ons avoid a second vendor
Trade-offs
- Streaming bills on connection time, not audio sent, which penalises idle sessions
- Add-ons are billed on top of the base rate
- Japanese accuracy needs verification
Editors' review
Watch the streaming billing model
Billing streaming on how long the WebSocket stays open rather than how much audio arrives is a meaningful difference. An agent holding a connection open through long silences pays for that silence.
PII redaction being native matters more than it sounds. Doing it after transcription means the unredacted text existed somewhere.
Verdict
Strong where redaction and understanding matter. Model streaming cost on real session lengths.
Compiled from official documentation and public sources, with our own editorial judgement. We do not publish figures we cannot source.