What is AssemblyAI?
AssemblyAI (Y Combinator Summer 2017) sells an API that turns audio into text. You send a recording or open a live stream and get back a transcript, with optional extras such as speaker labels, PII redaction, translation and a medical mode. Its current models are Universal-3.5 Pro for recordings and Universal-3.6 Pro Realtime for live audio; since Universal-3 Pro in February 2026 you can steer them with plain-language prompts and lists of key terms.
Pricing starts at $0.15 an hour for Universal-2 and $0.21 for Universal-3.5 Pro, with $50 of free credit and no card required. It also sells a Voice Agent API at $4.50 an hour and an LLM gateway. Its last confirmed round was a $50M Series C led by Accel in December 2023, bringing total funding to $115M.
How much does AssemblyAI cost?
| Plan | Price | What you get |
|---|---|---|
| Free credit | $50 one-off | No card; 5 new streams per minute |
| Pay as you go | From $0.15/hr | 100 new streams per minute |
| Voice Agent API | $4.50/hr | Speech-to-text, LLM and text-to-speech in one rate |
| Enterprise | Custom |
Pre-recorded: Universal-3.5 Pro $0.21/hr, Universal-2 $0.15/hr. Streaming: Universal-3.6 Pro Realtime $0.45/hr, Universal-Streaming $0.15/hr. Voice Agent API $4.50/hr. Add-ons from $0.01/hr; Medical Mode +$0.15/hr. $50 free credit.
Is AssemblyAI worth it?
Transcripts go wrong in the same places every time: product names, people's names, and the moment two speakers talk at once.
AssemblyAI lets you tell the model about those in advance. Since Universal-3 Pro in February 2026, you can pass a prompt and up to 1,000 key terms with each file, so the model knows your drug names or your customers' surnames before it hears them. On AssemblyAI's own benchmarks, Universal-3 Pro cut average English word error rate to 5.6% from 6.1% for Universal-2; there is no public model card, so treat those numbers as the vendor's.
The prices are low and published. An hour of recorded audio costs $0.21 on the best model; speaker labels add $0.02 and medical mode $0.15. A HIPAA agreement comes without a sales call at no extra cost, an EU region costs the same as the US, and SOC 2 Type 2 and ISO 27001 are included. Streaming is billed for as long as the connection stays open, so an idle stream still costs money.
The catch: by default AssemblyAI may use your audio to train its models, after redacting personal details. Paid customers can switch that off in the dashboard; free users cannot, and the switch only applies to new requests. Signing a BAA or using the EU servers turns training off for good. Recorded uploads are deleted within 24 to 48 hours and transcripts after 30 days unless you set a shorter time.
Our take: for most teams transcribing calls or meetings, AssemblyAI is the first price to beat, as long as someone flips the training switch on day one and keeps sensitive audio off the free tier.
Where it earns its place
A sales-call analytics startup processing 10,000 hours a month
Transcribe calls with speaker labels at a predictable cost
Universal-3.5 Pro at $0.21 plus standard diarization at $0.02 is $0.23 an hour, about $2,300 a month. Opt out of training in Data Controls before going live; stereo recordings cost double because each channel is billed.
A clinical documentation team at a 50-doctor practice group
Turn consultations into accurate notes in English and Spanish
Medical Mode adds $0.15 an hour; with diarization AssemblyAI's own example totals $0.38 an hour. The BAA is self-serve and, once signed, permanently opts the account out of training.
A voice agent developer
Get live transcripts fast enough for a phone bot
Universal-3.6 Pro Realtime at $0.45 an hour plugs into LiveKit, Pipecat, Vapi and Twilio. Close idle streams quickly, because billing runs for as long as the socket is open.
Before you commit
- Setup
- Sign up for $50 of free credit with no card, call the REST API or SDK with a file URL, and poll or receive a webhook for the transcript.
- Integrations
- REST and WebSocket APIs; integrations with LiveKit, Pipecat, Vapi and Twilio; an LLM gateway to OpenAI, Anthropic, Google and Qwen models at listed per-token prices.
- Data and privacy
- Training on customer audio by default after PII redaction; paid accounts can opt out in the dashboard, BAA and EU accounts are always opted out. Uploads deleted within 24 to 48 hours, transcripts after 30 days by default, zero retention for streaming once opted out.
- Cost at scale
- 10,000 hours of recorded calls a month is $2,100 on Universal-3.5 Pro before add-ons; the Voice Agent API at $4.50 an hour works out at $0.075 a minute.
- Limits
- 5 new streams per minute on the free tier and 100 on pay as you go; higher limits need an enterprise contract.
What's good
- Promptable models with up to 1,000 key terms per file
- $0.21 an hour for its best recorded-audio model
- Free HIPAA BAA without a sales call; EU region at US prices
- SOC 2 Type 2, ISO 27001 and GDPR at no premium
- $50 free credit with no card
What's not
- Customer audio used for training by default
- Free users cannot opt out of training
- Streaming billed for open connection time, including silence
- Accuracy figures are AssemblyAI's own, with no public model card
- Multichannel audio is billed per channel
Best for
Developers adding transcription or live captions to an app who want low published prices, a free BAA and an EU option.
AssemblyAI alternatives
AssemblyAI: common questions
- Is AssemblyAI worth it?
- AssemblyAI is cheap, accurate and easy to sign for, but read the data terms first: free accounts help train its models and cannot say no. We rate it 4.4 out of 5.
- How much does AssemblyAI cost?
- Pre-recorded: Universal-3.5 Pro $0.21/hr, Universal-2 $0.15/hr. Streaming: Universal-3.6 Pro Realtime $0.45/hr, Universal-Streaming $0.15/hr. Voice Agent API $4.50/hr. Add-ons from $0.01/hr; Medical Mode +$0.15/hr. $50 free credit.
- What are the downsides of AssemblyAI?
- Customer audio used for training by default; free users cannot opt out of training; streaming billed for open connection time, including silence; accuracy figures are AssemblyAI's own, with no public model card; multichannel audio is billed per channel.
- Who is AssemblyAI best for?
- Developers adding transcription or live captions to an app who want low published prices, a free BAA and an EU option.
- How much does AssemblyAI cost for a team or in production?
- 10,000 hours of recorded calls a month is $2,100 on Universal-3.5 Pro before add-ons; the Voice Agent API at $4.50 an hour works out at $0.075 a minute.
- Is AssemblyAI safe for company data?
- Training on customer audio by default after PII redaction; paid accounts can opt out in the dashboard, BAA and EU accounts are always opted out. Uploads deleted within 24 to 48 hours, transcripts after 30 days by default, zero retention for streaming once opted out.
- What are the best AssemblyAI alternatives?
- The closest alternatives we have reviewed are ElevenLabs, Recall.ai and Vapi.
Answers checked Oct 26, 2026.
Make AssemblyAI? Show this rating on your site
