AssemblyAI
assemblyai.com
- Free plan
- Web
- API
- Self-hosted
What AssemblyAI is
AssemblyAI provides speech-to-text, speech understanding and voice application models through APIs. Its Speech Understanding API can identify and separate speakers, summarize speech, extract action items and key phrases, detect sentiment, entities and topics, format and identify language, translate, redact personally identifiable information and moderate content. AssemblyAI lists support for 99 languages and translation into 86. It also states that streaming and voice-agent latency is below 300ms and that its platform processes more than 800 million API calls per month. Use the managed cloud or self-host models in your own environment. The service offers speaker labels, timestamps and SRT and VTT exports, with Python and JavaScript SDK guidance. Listed integrations include Twilio, Zoom RTMS, Zapier, LangChain and Vercel AI SDK. Security provisions include TLS 1.2+ in transit, AES-256 at rest, zero data retention and an option to opt out of model training. A free tier and usage-based plans are available; listed hourly rates include $0.15 per hour for Universal-2 and $0.45 per hour for Universal-3.6 Pro Realtime.
Who it suits
It suits developers building transcription, speech analysis or voice applications, including teams that need self-hosting or multilingual coverage. A free tier is listed for those starting with limited usage.
Good to know, and what to watch
- Supports transcription across 99 languages
- Includes speaker identification and timestamps
- Offers managed cloud or self-hosted deployment
- Provides Python and JavaScript SDK guidance
- Includes zero data retention and training opt-out
- Streaming is English-only on the listed Universal-Streaming plan
- Free tier limits new streaming connections to five per minute
- Universal-3.6 Pro Realtime is listed at $0.45 per hour
Plans
Every plan and priceFree credits
Pay as you go — Universal-2
Pay as you go — Universal-3.5 Pro
Pay as you go — Universal-Streaming
Getting started
- Open the AssemblyAI website
- Create an account for the free tier or a pay-as-you-go plan
- Use the API with the provided documentation
- Choose managed cloud or self-host the models
- Connect an official integration or use Python and JavaScript SDK guidance
- Send audio for transcription or speech understanding
Limits to check
The free tier allows up to 185 hours of pre-recorded transcription, up to 333 hours of streaming transcription, and 5 new streaming connections per minute. Free credits also include $50 in audio credits, with 5 concurrent pre-recorded transcriptions.
The Club's review
AssemblyAI is a developer-focused speech AI platform for adding transcription, speech understanding, and voice features to products. It is best suited to teams that need API access, real-time processing, or the option to run models in their own environment. The central decision is which model and billing tier match your languages, latency needs, and volume.
Overview
AssemblyAI combines speech-to-text with analysis that can turn conversations into structured data. Its managed cloud reduces infrastructure work, while self-hosting gives organizations control over deployment. Usage-based billing has no minimum commitment, upfront fee, contract, or required monthly subscription; invoices are issued monthly for the previous usage.
The platform says it handles more than 800 million API calls per month and targets sub-300ms latency for streaming and voice agents. Those claims make it a plausible fit for production applications, although your practical cost and language coverage depend on the selected model.
Key features
The Speech Understanding API supports speaker diarization and identification, summaries, action-item extraction, sentiment, key phrases, entities, topics, formatting, language detection, translation, personally identifiable information redaction, and content moderation. That breadth is valuable when a transcript must feed search, workflow, or analytics; a basic captioning project may not need so many processing layers.
Transcription coverage reaches 99 languages and translation reaches 86. Coverage is model-specific: Universal-2 supports 99 languages, Universal-3.5 Pro supports 18, and Universal-3.6 Pro Realtime supports 32. Universal-Streaming is English-only, while Universal-Streaming Multilingual supports English, Spanish, German, French, Portuguese, and Italian. Speaker labels, timestamps, SRT and VTT export, and API access cover common downstream requirements.
Official integrations include LiveKit, Pipecat, OpenRouter, Zapier, Make, n8n, Postman, Recall.ai, Zoom RTMS, Telnyx, Twilio, Amazon Connect, Genesys Cloud, LangChain, Vercel AI SDK, Power Automate, Semantic Kernel, Cloudflare, Bubble, Pipedream, and Drupal. Python and JavaScript SDK guidance, API documentation, cookbooks, support resources, a changelog, and a service-status page give developers several ways to build and maintain integrations.
Security controls include TLS 1.2 or newer in transit, AES-256 at rest, zero data retention, and an option to exclude data from model training. AssemblyAI maintains an annually audited SOC 2 Type 2 report and performs annual third-party penetration tests. Self-hosting is available for customers that need models inside their own environment.
Pricing
AssemblyAI has a free tier and pay-as-you-go plans. Audio usage is billed monthly, with pay-as-you-go audio prorated to the second where stated; there is no subscription commitment.
| Plan | Price and allowance | Best fit and trade-off |
|---|---|---|
| Free credits | Free; $50 in audio credits, 5 new streaming connections per minute, 5 concurrent pre-recorded transcriptions | Evaluation and prototypes; connection and concurrency limits constrain live or batch workloads. |
| Free tier | Free; up to 185 hours pre-recorded, up to 333 hours streaming, 5 new streaming connections per minute | Substantial experimentation before paying, but streaming connection creation remains capped. |
| Universal-2 | $0.15 a month, billed monthly on actual usage; 99 languages; 200+ concurrent pre-recorded transcriptions | The broad-language, lower-price speech-to-text choice for batch work. |
| Universal-3.5 Pro | $0.21 a month, billed monthly on actual usage; 18 languages; 200+ concurrent pre-recorded transcriptions | Higher-priced model option when its narrower language set is sufficient; custom rate limits are available. |
| Universal-Streaming | $0.15 a month, billed monthly on actual usage; 5 new streams per minute on free accounts and 100+ starting sessions per minute on paid accounts | English realtime transcription with a much higher paid-account session rate. |
| Universal-3.5 Pro Realtime | $0.45 a month, billed monthly on actual usage; 18 languages; 100+ starting sessions per minute on paid accounts | Realtime workloads needing the Pro option; it costs three times the listed Universal-Streaming rate. |
| Universal-Streaming Multilingual | $0.15 a month, billed per hour; English, Spanish, German, French, Portuguese, and Italian | Multilingual realtime in six named languages. |
| Universal-3.6 Pro Realtime | $0.45 a month, billed per hour; 32 languages, context carryover, and conversation memory | Realtime conversations that benefit from retained context. |
| Voice Agent API | $4.50 a month, billed per hour of connected conversation time; managed orchestration and hosting with no per-layer add-ons or concurrency fees | Teams wanting hosted voice-agent operations rather than assembling each layer themselves. |
Platforms
AssemblyAI is offered through API, web, and self-hosted deployment. That combination suits developers who want a managed starting point but may later require models inside a controlled environment.
Who it's for
Choose AssemblyAI for applications that need transcription plus speaker-aware analysis, redaction, moderation, summaries, or voice-agent behavior. It is also a strong fit for multilingual products using Universal-2, or for teams that need high concurrency and documented integrations. Look elsewhere if your project is a one-off transcription task, requires languages outside a chosen model's coverage, or needs a simple consumer-facing editor rather than APIs.
Pros and cons
Pros
- Speech Understanding adds diarization, sentiment, entities, topics, summaries, and compliance-oriented processing beyond raw transcripts.
- 99-language transcription, self-hosting, and a large integration catalog support varied production architectures.
- Usage billing without contracts or minimums limits financial commitment while a project is scaling.
- TLS 1.2+, AES-256, zero retention, model-training opt-out, SOC 2 Type 2 auditing, and annual penetration tests address sensitive audio concerns.
Cons
- Language counts vary sharply by model, with some realtime options limited to English or six languages.
- The free allowances impose streaming-connection and concurrency caps, so they are not a substitute for production capacity.
- Pro realtime and voice-agent rates are materially higher than the basic transcription tiers.
- Teams wanting only straightforward transcription may pay for capabilities they do not use.
Alternatives
For a broader comparison of categories, see Podcast Transcription Software, Audio Transcription Software, Speech Recognition Software, Speech-to-Text Software, and Transcription Software.
Simon Says is a freemium alternative with extension, desktop, mobile, web, and self-hosted platforms and a $15-per-hour pay-as-you-go plan. Happy Scribe is a freemium option with Android, iOS, macOS, and API support. Whipscribe offers a free first transcript and 99-plus-language auto-detection. OpenWhispr provides local-model support and a free weekly transcription allowance across desktop platforms. Rev is an English-focused freemium service with 45 AI transcription and caption minutes per month on its free plan. AudioPod AI is aimed at audio creation with credits for transcription, text-to-speech, and music. Speechmatics offers API, web, desktop, and self-hosted deployment with free credits and no free trial. Trint is a paid platform with a free trial and broad desktop, mobile, web, and API access.
Verdict
AssemblyAI is a strong choice for developers building speech products that need analysis as well as transcription, especially when self-hosting, security controls, integrations, or usage billing matter. Choose a different service when your workflow is a simple consumer transcription job or when the required languages and realtime model limits do not align with AssemblyAI's tiers.
The verdict
AssemblyAI combines transcription with a broad set of speech-analysis APIs and offers both managed and self-hosted deployment. Check the language coverage and usage limits of the specific plan before choosing it.
Questions people ask
Is AssemblyAI free?
A free tier is available, including free audio credits and stated transcription and connection allowances.
How much do paid plans cost?
Listed rates include $0.15 per hour for Universal-2, $0.21 per hour for Universal-3.5 Pro, $0.45 per hour for Universal-3.6 Pro Realtime, and $4.50 per hour for the Voice Agent API.
How is usage billed?
Pay-as-you-go usage is billed monthly based on actual use, with no minimum commitments, upfront fees, contracts, or monthly subscription requirement.
Can AssemblyAI be self-hosted?
Yes. Customers can run it in the managed cloud or self-host models inside their own environment.
What languages does it support?
The platform states coverage across 99 languages and translation into 86 languages.
Which export formats are available?
SRT and VTT export formats are listed.
What the maker publishes
- API access
- Yes source
- Billing
- Pay-as-you-go billing has no minimum commitments, upfront fees, contracts, or monthly subscription requirement, and invoices are generated monthly for prior usage. source
- Compliance
- AssemblyAI maintains an annually audited SOC 2 Type 2 report and conducts annual third-party penetration tests. source
- Deployment
- Customers can run AssemblyAI in its managed cloud or self-host the models inside their own environment. source
- Developer support
- AssemblyAI provides documentation, an API reference, cookbooks, support resources, a changelog, and a service-status page. source
- Export formats
- SRT,VTT source
- Free plan
- No source
- Integrations
- Official integrations include LiveKit, Pipecat, OpenRouter, Zapier, Make, n8n, Postman, Recall.ai, Zoom RTMS, Telnyx, Twilio, Amazon Connect, Genesys Cloud, LangChain, Vercel AI SDK, Power Automate, Semantic Kernel, Cloudflare, Bubble, Pipedream, and Drupal. source
- Languages supported
- 99 source
- Language coverage
- The platform states coverage across 99 languages and translation into 86 languages. source
- Latency
- AssemblyAI states sub-300ms latency for streaming and voice agents. source
- Scale
- AssemblyAI states that its platform processes more than 800 million API calls per month. source
- SDKs
- Official documentation provides Python and JavaScript SDK guidance. source
- Security
- AssemblyAI offers TLS 1.2+ encryption in transit, AES-256 encryption at rest, zero data retention, and an opt-out from model training. source
- Speaker identification
- Yes source
- Speech features
- Its Speech Understanding API supports speaker diarization, speaker identification, summarization, action items, sentiment analysis, key phrases, entity detection, topic detection, formatting, language detection, translation, PII redaction, and content moderation. source
- Timestamp support
- Yes source
- What it does
- AssemblyAI provides Voice AI models and APIs for speech-to-text, speech understanding, and voice applications. source
Checked against the maker's pages on 1 Oct 2026.