Quick verdict
Bottom line up top: AssemblyAI is a developer API, not an app — and judged as an API, it is one of the cleanest deals in speech-to-text: pre-recorded transcription from $0.15 an hour, a $50 free credit to start, and no subscription or minimums. The catch is the add-on menu: diarization, redaction, summarization and other extras each bill per hour on top, so your real rate is the base rate plus whatever your product actually needs.
Research-based review — no hands-on testing claimed. Prices checked October 24, 2026.
What does AssemblyAI do?
AssemblyAI turns audio into text — and then into structured insight — through an API that developers plug into their own products. Its current models include Universal-2 (broad coverage, 99+ languages) and the Universal-3.5 Pro line (its accuracy flagship, with native code-switching across 18 languages), plus real-time streaming models for live captions and voice agents over WebSocket.
Around the transcription core sits an Audio Intelligence layer: speaker diarization (who spoke when), sentiment analysis, entity and topic detection, summarization, content moderation, and PII redaction. An LLM Gateway lets you run models such as Claude or Gemini over the transcript, and a Voice Agent API bundles transcription, reasoning and turn detection at a flat hourly rate for teams building conversational agents.
How much does AssemblyAI cost in 2026?
| Item | Price | Notes |
|---|---|---|
| Free tier | $50 in credits, no card required | Roughly 185 hours of pre-recorded transcription at base rates |
| Universal-2 (pre-recorded) | $0.15/hr | 99+ languages; the budget workhorse |
| Universal-3.5 Pro (pre-recorded) | $0.21/hr | Accuracy flagship; native code-switching in 18 languages |
| Streaming | from $0.15/hr; $0.45/hr for the top realtime model | Live captions and voice-agent input |
| Voice Agent API | $4.50/hr flat | Transcription + reasoning + speech bundled |
| Enterprise | Custom | Volume pricing, higher concurrency, dedicated support; HIPAA/SOC 2/ISO 27001/GDPR coverage |
Everything is billed per second of audio with no minimums. Add-ons are priced separately per hour — speaker diarization, PII redaction, Medical Mode and prompting among them — and they only bill when you use them. That is fair pricing, but it means a feature-rich pipeline can cost several times the headline $0.15.
What are AssemblyAI’s strengths?
- Headline rates are genuinely low. $0.15–$0.21 an hour undercuts per-minute rivals by a wide margin at volume.
- A real free on-ramp. $50 of credit with no card is enough to transcribe a serious evaluation set on your own audio.
- One API for the whole audio stack. Transcription, diarization, summaries, redaction and LLM queries without stitching four vendors together.
- Language breadth with a flagship path. Universal-2 covers 99+ languages; the Pro line handles code-switching conversations where speakers mix languages mid-sentence.
- Compliance story for regulated buyers. Enterprise covers HIPAA, SOC 2 Type 2, ISO 27001 and GDPR requirements.
What are AssemblyAI’s weaknesses?
- It is API-only. There is no friendly app for a podcaster or a meeting host — someone on your team has to write and maintain code.
- Add-on math. The per-hour extras stack; budget from a worked example of your real pipeline, not the base rate.
- Flagship language coverage is narrower. The best accuracy and code-switching cover 18 languages; long-tail languages fall back to Universal-2.
- Accuracy still needs your own test. Vendor accuracy claims are made on vendor audio — noisy calls, accents and domain jargon are exactly what the free credit is for.
Who is AssemblyAI for?
Product and platform teams building transcription, captions, call analytics, meeting intelligence or voice agents into their own software — especially at volumes where per-minute pricing elsewhere becomes the budget. Solo creators who just want transcripts should buy an app (Otter, Sonix and friends) instead of an API.
Frequently asked questions
Is AssemblyAI free?
New accounts get $50 in credits with no credit card — roughly 185 hours of pre-recorded transcription at the $0.15–$0.21 base rates, less once add-ons are switched on. After that it is pay-as-you-go.
How much does AssemblyAI cost per hour?
Pre-recorded transcription is $0.15/hr on Universal-2 and $0.21/hr on Universal-3.5 Pro. Streaming starts at $0.15/hr and the top realtime model is $0.45/hr. Add-ons bill extra per hour.
Does AssemblyAI work in real time?
Yes — streaming models transcribe over WebSocket for live captions and voice agents, and the Voice Agent API bundles the full conversational loop at $4.50/hr.
AssemblyAI vs Deepgram: which should I pick?
Both are strong developer STT APIs. AssemblyAI leans into its bundled Audio Intelligence layer and very low async rates; Deepgram leans into streaming speed and per-minute pricing. Run the same hour of your own audio through both free tiers and compare error rates and total bills.
The verdict
If you are building audio into a product, AssemblyAI belongs on your shortlist and its free credit makes the evaluation nearly free. Just price the whole pipeline — base model plus every add-on you will actually enable — before you fall in love with the $0.15 headline.