by AssemblyAI, Inc.Speech-to-text APIs
A speech-to-text API priced per audio hour, from $0.15 for Universal-2 batch or Universal-Streaming, with $50 of free credit and a self-serve BAA on paid accounts.
A transcription API whose BAA, training opt-out and retention controls are self-serve on paid accounts, with diarized batch audio from $0.17 an hour.
Paid accounts can sign a HIPAA BAA online at no cost, opt out of model training, and set how long audio and transcripts are kept, down to one hour, without a sales contract. The DPA applies with the terms, and the EU region costs the same as the US. Universal-2 transcribes 99 languages in batch for $0.15 an hour plus $0.02 for speaker labels, and add-ons such as PII redaction and keyterm prompting have per-hour prices.
Against 3 other products we reviewed in Speech-to-text APIs: what you pay for with each, the five access facts, the plan limits we recorded and each review's bottom line. The note under the table says how each one was checked.
| Metric | AssemblyAI | Speechmatics | Google Cloud Speech-to-Text | Amazon Transcribe |
|---|---|---|---|---|
| What you're paying for | Compliance includedA transcription API whose BAA, training opt-out and retention controls are self-serve on paid accounts, with diarized batch audio from $0.17 an hour. | Compliance includedTranscription in 55+ languages where your audio stays out of model training unless you opt in, with EU, US and Australian processing for every account. | Built-in reachSpeech recognition inside your Google Cloud account, with IAM, audit logs, a BAA and EU processing, and dynamic batch at $0.003 a minute. | Built-in reachTranscription inside your AWS account at $0.006 a minute for batch, with speaker labels, custom vocabulary and language identification in the rate. |
| Overall score | 8.4Best | 8.1 | 7.8 | 7.0 |
| Free option | Free trial | Free trial | Free plan | Free trial |
| Card to start | No card needed | No card needed | Card required | –Not measured |
| Getting access | Self-serve | Self-serve | Self-serve | Self-serve |
| Starter priceCheapest paid plan | $0.15/hr | $0.24/hr | $0.003/min | $0.006/min |
| Pricing model | Usage-based | Usage-based | Usage-based | Usage-based |
| Pricing page | Public prices | Public prices | Public prices | Public prices |
| Billing unit | Per second of pre-recorded audio; streaming by session time, idle included; multichannel per channel | Per second, from credits at hourly rates; credit price falls with packs and subscriptions | Per second of audio, no minimum; each channel billed | Per second, no minimum; two channels at no extra charge |
| Free credit | $50 at sign-up, no card, does not expire | $100 at sign-up, no card, valid 12 months | 60 minutes a month on the V1 API; none on V2 | 60 minutes a month for 12 months |
| Streaming rate | Universal-Streaming $0.15/h; Universal-3.6 Pro Realtime $0.45/h | Melia 1 $0.40/h; Standard $0.45/h; Enhanced $0.80/h (list) | $0.016/min ($0.96/h), the same as batch | $0.01/min ($0.60/h) |
| Diarization | Batch +$0.02/h; streaming +$0.12/h (beta in the docs) | Included; real-time speaker labels on Enhanced, Standard and Linden, not Melia 1 | No separate price; Chirp 3 diarization in batch only | Included in batch and streaming, up to 30 speakers |
| Languages | 99 on Universal-2; 18 on Universal-3.5 Pro | 55+ (pricing page); 56 for Melia 1 | Over 125 (product page) | About 90 languages across 113 language rows (our count) |
| Training on your audio | By default; paid accounts opt out online, free accounts cannot | Only if you opt in (33% off) | Not without your permission; V1 data logging opt-in gives a lower rate | By default; opt out with an AWS Organizations policy |
| HIPAA BAA | Signed online at no cost on paid plans | Sub-BAA through the Trust Center; plan and process not stated | Covered; accepted in the console per the docs | Accepted in AWS Artifact |
| EU processing | EU endpoint for batch and streaming, same price as US | EU, US and Australia regions for all customers | Chirp 3 GA in the eu multi-region | Six EU regions for batch, four for streaming |
| Self-hosted | Streaming only, $20,000 upfront commitment | On-prem containers, virtual appliance and on-device on Enterprise | Speech-to-Text On-Prem on GKE, through sales | Not offered |
| SOC 2 / ISO 27001 | SOC 2 Type 2; ISO 27001:2022 | SOC 2 Type II (Trust Center, under NDA); ISO/IEC 27001:2022 | SOC 2 Type II and ISO 27001 (Compliance Reports Manager) | SOC 1, 2 and 3 and ISO 27001 (AWS Artifact) |
| Console SSO | SAML/OIDC add-on, $199/month per connection | SAML/OIDC on Enterprise, $280/month add-on | SAML through Cloud Identity Free, no listed cost | IAM Identity Center, no extra charge |
| Bottom line | From its public pages, AssemblyAI combines the lowest 500-hour price here with compliance controls you switch on yourself: a BAA, a training opt-out and retention as short as an hour. A few add-on prices differ between its pages. | From its public pages, Speechmatics has the strictest privacy defaults of the five: no training unless you opt in, 7-day batch retention and real-time audio not stored. Its pricing page leads with discounted rates, and redaction, SSO and on-prem are Enterprise features. | From its public pages, Google Cloud Speech-to-Text suits teams inside Google Cloud: no training without permission, a BAA, EU processing and dynamic batch at $0.003 a minute. Features need checking per model, and there is no PII redaction. | From its public pages, Amazon Transcribe is the simple choice inside AWS: flat per-minute rates with the main features included and self-serve compliance paperwork. Voice inputs are used to improve the service unless your organization opts out. |
Evidence: AssemblyAI: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run; Speechmatics: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run; Google Cloud Speech-to-Text: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run; Amazon Transcribe: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run. Prices in USD as listed, excluding tax. "–" means not recorded.
Review from public pages
Written from AssemblyAI's public pages (pricing, security or privacy, and docs) as read on Oct 16, 2026. We did not sign up, and no hands-on run is published yet.
We read AssemblyAI's pricing page and the pricing.md and models.md files it links from the footer, the billing, rate-limit, data retention, Data Controls, region, SSO, account management and self-hosting docs, the docs FAQ, the security page, the Terms of Service, the DPA and the BAA, and the changelog on Oct 16, 2026; we did not sign up or call the API for this review. The trust center did not render without JavaScript, so certifications come from the security page and the docs.
Pricing is per audio hour, model by model, with every add-on listed at its own hourly rate, and the billing docs explain the units well: pre-recorded audio to the second, streaming by how long the WebSocket stays open, idle time included, and each channel billed separately.
Several pages disagree on details. The pricing page lists Universal-2 keyterm prompting as included while pricing.md lists +$0.05 an hour; the Sync API is $0.30 an hour on the pricing page and $0.45 in the changelog; the FAQ says the $50 credit covers up to 185 batch hours; and a docs FAQ says SAML SSO is not supported, while the SSO page offers it at $199 a month.
Paid accounts sign the BAA online, opt out of training and set retention in Data Controls; the DPA applies automatically with the terms.
vs Deepgram: Deepgram's BAA is for Enterprise customers, and its DPAs are entered into on request.
Universal-2 batch at $0.15 an hour plus $0.02 for diarization puts 500 hours at $85 a month (computed).
vs Speechmatics: Speechmatics Melia 1 comes to $120 and Deepgram Nova-3 to $129 for the same scenario.
The security page lists a SOC 2 Type 2 report in the trust center, and the docs state ISO 27001:2022 and PCI DSS v4.0.
vs Deepgram: Deepgram does not state ISO 27001.
Diarization, keyterm prompting, PII redaction, Medical Mode, translation and the Speech Understanding models each have an hourly price, so a feature set can be costed from the page.
The Terms let AssemblyAI use customer data to train its models; paid accounts can opt out, forward-looking only, and free accounts cannot.
Workaround: Per the data retention docs, files are not used for training under a BAA, on the EU servers, or after an opt-out.
Streaming bills the time the WebSocket stays open, idle time included, and an unclosed session runs to the 3-hour cap; add-ons such as diarization apply to the whole session.
Workaround: Close each session when the audio ends, as the billing docs advise.
Universal-2 keyterm prompting, the Sync API and summarization each have different prices on the pricing page and in pricing.md or the changelog.
Workaround: Confirm the rates for the add-ons you need with AssemblyAI.
The newer Universal-3.5 Pro lists 18 batch languages; the 99-language model is Universal-2. Streaming diarization is still described as beta in the docs.
AssemblyAI draws usage from a prepaid balance, pre-recorded audio to the second and streaming by session time. For our scenario, 500 hours of pre-recorded English with diarization and word timestamps on Universal-2 cost 500 × ($0.15 + $0.02) = $85 a month (computed) before tax; on Universal-3.5 Pro it would be $115. PII redaction adds $0.08 an hour ($40 a month at that volume, computed) and Medical Mode $0.15. Streaming starts at $0.15 an hour on Universal-Streaming plus $0.12 for diarization, with idle session time billed. Volume pricing is through sales, and SAML SSO is $199 a month per connection. Paid fees are not refundable under the Terms.
Transcripts are available as JSON and as SRT or VTT captions at no charge. GET /v2/transcript lists transcripts from the last 30 days only, and transcripts are deleted on the retention schedule, so export what you need to keep. Account deletion is permanent, and the free credit does not expire.
From its public pages, AssemblyAI combines the lowest 500-hour price here with compliance controls you switch on yourself: a BAA, a training opt-out and retention as short as an hour. A few add-on prices differ between its pages.
| Limit | As recorded |
|---|---|
| Billing unit | Per second of pre-recorded audio; streaming by session time, idle included; multichannel per channel |
| Free credit | $50 at sign-up, no card, does not expire |
| Streaming rate | Universal-Streaming $0.15/h; Universal-3.6 Pro Realtime $0.45/h |
| Diarization | Batch +$0.02/h; streaming +$0.12/h (beta in the docs) |
| Languages | 99 on Universal-2; 18 on Universal-3.5 Pro |
| Training on your audio | By default; paid accounts opt out online, free accounts cannot |
| HIPAA BAA | Signed online at no cost on paid plans |
| EU processing | EU endpoint for batch and streaming, same price as US |
| Self-hosted | Streaming only, $20,000 upfront commitment |
| SOC 2 / ISO 27001 | SOC 2 Type 2; ISO 27001:2022 |
| Console SSO | SAML/OIDC add-on, $199/month per connection |
Read from the vendor's public pages on Oct 16, 2026. No hands-on run is published yet. How we check
We have not published a sign-up run for AssemblyAI yet. The access facts above were read from its sign-up and pricing pages on Oct 16, 2026.
We have not published API calls for AssemblyAI yet.
AssemblyAI lists 6 plans. The cheapest paid plan, Universal-2, pre-recorded, costs $0.15/hr.
Prices as listed, USD, excluding tax; observed 2026-10-16.
Source: assemblyai.com pricing page. Prices change; the observed date above is when our agent read them.
We read the pricing, docs, data-usage, trust and legal pages of three transcription APIs on the same day and scored them on the same six indicators. What 500 hours of diarized batch audio cost, how each handles your audio, where the BAA sits, and who should pick which.
