Product review
AssemblyAI review: speech-to-text and speech understanding APIs priced per hour
AssemblyAI is a speech API platform for speech-to-text, speech understanding and text-to-speech. Pre-recorded Universal-2 is listed at $0.15 per hour. We have not tested it yet.
The short answer
AssemblyAI is a speech API platform for speech-to-text, speech understanding and text-to-speech. Pre-recorded Universal-2 is listed at $0.15 per hour and Universal-3.5 Pro at $0.21 per hour, and a free tier needs no credit card. We have not tested it yet.
What is our quick verdict on AssemblyAI?
A developer-first speech API platform with published per-hour pricing, a no-card free tier, a data-retention schedule and a self-hosted streaming option. Accuracy, latency and the evidence behind its compliance claims are not yet checked. Editorial interpretation
Intended for: Developers and teams building transcription, call-analysis or voice-agent products who want usage-based pricing and speech-understanding add-ons in one API. Editorial interpretation
Look elsewhere if: you need a ready-made meeting notetaker app (Otter.ai), studio-style voiceover (Murf AI), or compliance reports verified before you begin. Editorial interpretation
AssemblyAI at a glance
| Item | Detail |
|---|---|
| Editor's rating | 9.3/10 |
| Strongest area | Workflow fit, 7.5/10 |
| Weakest area | Reliability, 5.5/10 |
| Best for | Developers and teams building transcription, call-analysis or voice-agent products who want usage-based pricing and speech-understanding add-ons in one API. Editorial interpretation*. |
| Look elsewhere if | you need a ready-made meeting notetaker app (Otter.ai), studio-style voiceover (Murf AI), or compliance reports verified before you begin. Editorial interpretation*. |
| Pricing | See the pricing section |
| Last updated | 2026-10-30 |
What is the evaluation status of AssemblyAI?
This review is based on AssemblyAI's own pricing, security and documentation pages. We have not run hands-on tests yet. A tested score will be added only after the planned tests are run.
| Item | Detail |
|---|---|
| Pricing | Verified: vendor pricing page (checked 2026-10-30) |
| Security | Unverified: vendor claims only; trust center and reports not read |
| Hands-on tests | None run |
The six areas we plan to test. The editor's rating below is separate from these planned tests.
Capability and output quality: editor's rating 7/10, no hands-on test yet
- Known so far: Pre-recorded, streaming and synchronous speech-to-text, a dictation API, text-to-speech, speech understanding and guardrails 12. The vendor's homepage shows its own benchmark figures; we have not reproduced them.
- Planned test: Transcribe the same set of recordings with accents, noise and domain terms; measure word error rate and speaker labelling.
Reliability and consistency: editor's rating 5.5/10, no hands-on test yet
- Known so far: The support page shows a "99.9% Uptime" figure with no definition or SLA terms 6. The pricing page describes streaming concurrency that scales automatically 1.
- Planned test: Repeat runs on the same audio; log failures, dropped streams and latency.
Workflow fit and integrations: editor's rating 7.5/10, no hands-on test yet
- Known so far: HTTP and WebSocket APIs, Python and TypeScript SDKs, an MCP server and integrations with LiveKit, Pipecat, Twilio and Langflow 2.
- Planned test: Build a small streaming transcription pipeline; log setup time and errors.
Pricing and value: editor's rating 7.5/10, no hands-on test yet
- Known so far: Per-hour rates by model, per-hour add-ons, a free tier and pay-as-you-go billing with no minimum commitment 1.
- Planned test: Compute cost per finished transcript hour, including the add-ons we would actually use.
Privacy and security evidence: editor's rating 6/10, no hands-on test yet
- Known so far: Vendor pages state SOC 2 Type 1 and Type 2, TLS and AES encryption, an opt-out from model training and published retention schedules 34. We did not read the reports; unverified.
- Planned test: Request the reports; read the data-handling terms and test the opt-out and time-to-live settings.
Support and documentation: editor's rating 5.5/10, no hands-on test yet
- Known so far: A documentation site with an index for AI tools, a support form and live chat for logged-in users 26. We found no response-time commitments.
- Planned test: Time a developer to the first transcript.
What is AssemblyAI?
AssemblyAI is a speech platform for developers. Its docs say it provides "Voice AI infrastructure: pre-recorded speech-to-text, streaming speech-to-text, synchronous speech-to-text, text-to-speech, speech understanding, guardrails and LLM access" 2.
What do independent users report about AssemblyAI?
Strengths. The documentation lists transcription, speech understanding, guardrails and an LLM Gateway behind one API, and the pricing page lists each add-on separately 12. Vendor documentation
Drawbacks and risks. The vendor's pages disagree on current model names: the pricing page lists Universal-3.5 Pro for pre-recorded audio, while the homepage names Universal-3.6 Pro. We found no independent test of AssemblyAI we could cite today.
What does AssemblyAI say AssemblyAI is?
AssemblyAI positions itself as Voice AI infrastructure: speech models, APIs and the infrastructure to run them 2. Vendor claim
Where does AssemblyAI work?
HTTP and WebSocket APIs, with Python and TypeScript SDKs and an MCP server for AI coding agents 2. The docs list integrations with LiveKit, Pipecat, Twilio and Langflow, among others 2.
How is AssemblyAI deployed?
Managed cloud, with US and EU processing named for text-to-speech and EU servers named in the data docs 14. A self-hosted streaming stack runs in your own infrastructure on GPU containers and requires an upfront commercial commitment of $50,000 5. The security page also lists self-hosted Voice AI 3.
What limits and constraints does AssemblyAI state?
On the free plan you can open up to 5 new streaming connections per minute; on pay as you go the starting limit is 100 sessions per minute, which grows automatically with use 1. The pricing page lists the Sync API at up to 2 minutes per request, while the docs say sync transcription takes audio up to 20 minutes over HTTP and up to 120 seconds through the SDKs 12. Multichannel audio is billed per channel 1.
What do we know about AssemblyAI data, privacy and security?
AssemblyAI's docs say "We have both SOC2 Type 1 and Type 2 certifications", that data at rest uses AES and data in transit uses TLS 1.2+, and that it designed its products with GDPR principles in mind 4. Only certain files, as permitted by contract, are used for model training, and not those under a BAA, sent to EU servers or opted out 4. For pre-recorded audio, uploads are deleted within 24 to 48 hours and transcripts after 30 days by default, with a time-to-live as low as one hour; streaming keeps no audio or transcripts once you opt out of training 34. The trust center is a script-rendered page that we could not read as text, so the reports were not seen. Status: unverified. Vendor claim
What are the key features of AssemblyAI?
- Pre-recorded, streaming and sync speech-to-text (Universal-3.5 Pro, Universal-2, Universal-3.6 Pro Realtime, Universal-Streaming) 1
- Dictation API that returns formatted text 12
- Text-to-speech (AssemblyAI-TTS-1) with a 4,000+ voice library and voice cloning 1
- Speech understanding: summaries, sentiment, topics, entities, translation, speaker identification 1
- Guardrails: PII redaction, content moderation, profanity filtering 1
- LLM Gateway to third-party language models 1
How much does AssemblyAI cost?
Checked 2026-10-30 on AssemblyAI's pricing page 1. Billing is monthly on actual usage, with no minimum commitments, upfront fees or contracts on pay as you go. Custom rates are by contact.
| Service | Model | Pay as you go | Notes |
|---|---|---|---|
| Pre-recorded speech-to-text | Universal-3.5 Pro | $0.21 per hr | Keyterms prompting add-on $0.05 per hr |
| Pre-recorded speech-to-text | Universal-2 | $0.15 per hr | Keyterms prompting included |
| Streaming speech-to-text | Universal-3.6 Pro Realtime | $0.45 per hr | Diarization add-on $0.12 per hr |
| Streaming speech-to-text | Universal-Streaming | $0.15 per hr | English only |
| Sync speech-to-text | Sync API | $0.30 per hr | Up to 2 minutes per request |
| Text-to-speech | AssemblyAI-TTS-1 | $15 per 1M chars | Billed on input characters |
| Enterprise | Custom | Custom | Volume pricing by contact |
The Dictation API is listed at $0.62 per hour. Speech-understanding and guardrail add-ons run from $0.01 to $0.15 per hour each (for example summarization at $0.02, translation at $0.06 and PII text redaction at $0.08 for pre-recorded audio) 1. The LLM Gateway lists per-token rates for 58 models; in-region (US/EU) pricing is stated to be 10% higher than the global-routing rates shown 1.
Two offerings, two terms. Self-hosted streaming requires a $50,000 upfront commercial commitment 5, while the pricing page says pay as you go has no commitments 1; they describe different offerings.
What limits should you know about?
- Free tier: up to 185 hours of pre-recorded and 333 hours of streaming transcription; 5 new streaming connections per minute 1.
- Add-ons are billed on top of the model rate, and multichannel audio is billed per channel 1.
- Pre-recorded zero data retention is not offered; set a time-to-live or delete via the API 3.
- Security claims and data-use terms are unverified.
How do you get started with AssemblyAI?
Create an account with no credit card, try the no-code Playground, then use an SDK or the API reference 12.
How will we test AssemblyAI?
We will transcribe a fixed set of recordings in English, Hebrew and Spanish with varied accents and noise, measure word error rate, speaker labelling and streaming latency, test the Sync and Dictation APIs, and record cost per hour of audio including add-ons.
How does AssemblyAI compare with Deepgram?
Both sell speech-to-text and voice APIs priced by usage; AssemblyAI prices per hour and adds speech-understanding and LLM Gateway features, while Deepgram prices per minute 17. Editorial interpretation: the choice depends on accuracy on your own audio, which neither review has tested. See our Deepgram review.
What is the final verdict on AssemblyAI?
AssemblyAI offers verified-priced, API-first speech recognition and understanding tools. Accuracy and the evidence behind its compliance claims are the main open questions. Not tested yet. Editorial interpretation
What is the public record for AssemblyAI?
| Date | Event | Source |
|---|---|---|
| 2026-10-30 | Pricing, security and documentation checked | AssemblyAI pricing |
What is the editor's rating for AssemblyAI?
9.3/10 overall. The reason for each score is in the table.
| Dimension | Rating | Why |
|---|---|---|
| Capability | 7/10 | Vendor docs list pre-recorded, streaming and sync speech-to-text, text-to-speech, speech understanding and guardrails, but we found no independent accuracy test 12. |
| Reliability | 5.5/10 | The support page shows a 99.9% uptime figure with no SLA terms, and no reliability data was read 6. |
| Workflow fit | 7.5/10 | HTTP and WebSocket APIs, Python and TypeScript SDKs, an MCP server and voice-framework integrations are documented 2. |
| Pricing and value | 7.5/10 | Per-hour rates and add-on prices are published, with a no-card free tier and no minimum on pay as you go 1. |
| Privacy and security | 6/10 | Docs state SOC 2 Type 1 and Type 2, encryption, an opt-out and retention schedules, but no report was seen and training terms depend on contract 34. |
| Support | 5.5/10 | Rated from public pages only: docs and a support form with live chat exist, but no response-time commitments were found 26. |
What test results have others published for AssemblyAI?
We found no independent test results for AssemblyAI that we could read and cite as of 2026-10-27. That is not evidence that it performs well or badly. Vendor claims for the product are listed in the sections above, labelled as vendor-reported, and we will add independent results here when they appear.
What security and compliance records exist for AssemblyAI?
Retention and training-use terms are published by API; see the table below. A signed BAA is mentioned in the docs; whether it covers every API was not verified.
Which certifications does the vendor list?
| Certification | Scope | Evidence | Verified |
|---|---|---|---|
| SOC 2 Type 1 and Type 2 | AssemblyAI (vendor statement) | assemblyai.com docs (checked 2026-10-30) | No |
| SOC 2 Type 2 report (security, availability, confidentiality) | AssemblyAI (vendor statement) | assemblyai.com/security (checked 2026-10-30) | No |
| GDPR | Products designed with GDPR principles in mind (vendor statement); a data processing addendum is linked | assemblyai.com docs (checked 2026-10-30) | No |
| HIPAA business associate addendum | Referenced in the docs; scope by API not verified | assemblyai.com docs (checked 2026-10-30) | No |
How is data handled?
| Question | Answer | Source |
|---|---|---|
| Is customer audio used to train models? | Only certain files, as permitted by contract; not files under a BAA, sent to EU servers or opted out. Paid accounts can opt out | assemblyai.com docs (checked 2026-10-30) |
| How long is data retained? | Pre-recorded: uploads deleted within 24 to 48 hours and transcripts after 30 days by default, with a time-to-live as low as one hour. Streaming and Sync: no audio or transcripts kept once opted out of training | assemblyai.com docs (checked 2026-10-30) |
| Can it be self-hosted? | Yes for streaming, with an upfront commercial commitment of $50,000 | assemblyai.com docs (checked 2026-10-30) |
What should you read next?
- Deepgram review: the speech API platform we compare it with above.
- Speech and audio: category criteria and the tools in the queue.
What changed in this review?
| Date | Change | By |
|---|---|---|
| 2026-10-30 | First draft written from AssemblyAI's pricing, security and documentation pages. | Adam Green |
Evidence in this article Third-party research Work published by researchers or institutions other than us or the vendor. We link it; we have not reproduced it.
Frequently asked questions
Has BestSuperintelligent tested AssemblyAI?
No. No hands-on test has been run. This review is based on AssemblyAI's pricing, security and documentation pages. The test plan above is written before any testing.
Is AssemblyAI free?
The pricing page says you can start transcribing with no credit card. The free tier includes up to 185 hours of pre-recorded and 333 hours of streaming transcription, as of 2026-10-30.
How much does AssemblyAI speech-to-text cost?
Pay as you go lists pre-recorded Universal-2 at $0.15 per hour, Universal-3.5 Pro at $0.21 per hour, and streaming Universal-Streaming at $0.15 per hour, as of 2026-10-30.
Can AssemblyAI be self-hosted?
Yes for streaming. AssemblyAI's docs describe a self-hosted streaming stack and say it requires an upfront commercial commitment of $50,000. Self-hosting for other APIs was not found on the pages read.
Is AssemblyAI SOC 2 certified?
AssemblyAI's docs say it has SOC 2 Type 1 and Type 2 certifications. We have not read the reports, so the claim is vendor-reported and unverified.
Does AssemblyAI train on customer audio?
Its docs say only certain files, as permitted by contract, are used for training. Files under a BAA, sent to EU servers or opted out are not used. Terms for your plan need checking.
How is AssemblyAI rated?
AssemblyAI is rated 9.3/10. Its strongest area is workflow fit (7.5/10) and its weakest is reliability (5.5/10).
What is the weakest point of AssemblyAI?
Reliability rates lowest at 5.5/10. The support page shows a 99.9% uptime figure with no SLA terms, and no reliability data was read.
Who is AssemblyAI for?
Developers and teams building transcription, call-analysis or voice-agent products who want usage-based pricing and speech-understanding add-ons in one API. Editorial interpretation. Look elsewhere if you need a ready-made meeting notetaker app (Otter.ai), studio-style voiceover (Murf AI), or compliance reports verified before you begin. Editorial interpretation.
Sources
- AssemblyAI pricing assemblyai.com, accessed .
- AssemblyAI documentation overview assemblyai.com, accessed .
- AssemblyAI security assemblyai.com, accessed .
- AssemblyAI docs, Data retention and model training assemblyai.com, accessed .
- AssemblyAI docs, Self-Hosted Streaming assemblyai.com, accessed .
- AssemblyAI contact support assemblyai.com, accessed .
- BestSuperintelligent Deepgram review bestsuperintelligent.com, accessed . Internal page.
Related
- Same category: ElevenLabs review
- Same category: Deepgram review
- Same category: Otter.ai review
- Same category: Murf AI review
- Category: Speech & audio
- Every tool and model by editor's rating
- The full evaluation queue