AssemblyAI is a speech-to-text and audio-intelligence API for developers, turning recorded or streaming audio into transcripts, speaker labels, summaries and other structured data. It's built for engineering teams shipping voice AI products, such as call analytics, meeting-notes tools, and voice agents, rather than for end users looking for a consumer app. Pricing is entirely usage-based: pre-recorded transcription starts at $0.15/hour on the Universal-2 model, real-time streaming starts at $0.15/hour, and new accounts get $50 in free credit with no credit card required.
Who builds it
AssemblyAI is built by AssemblyAI, Inc., founded in 2017 by CEO Dylan Fox and backed by a $50M Series C round led by Accel (announced December 2023) with participation from Y Combinator alumni investors, Insight Partners and others. The core workflow: send an audio file or a WebSocket stream to the API with a chosen model, and AssemblyAI returns a transcript plus optional add-on outputs like speaker diarization, sentiment analysis, summarization or PII redaction, which developers pull into their own product through the REST API or the official Python and JavaScript/Node.js SDKs.
Core features
- Pre-recorded speech-to-text via the Universal-3.5 Pro ($0.21/hr) and Universal-2 ($0.15/hr) models, the latter supporting 99 languages
- Real-time streaming transcription (Universal-3.5 Pro Realtime, Universal-Streaming English/Multilingual) billed on connection duration, from $0.15 to $0.45/hour
- Voice Agent API bundling speech-to-text, LLM reasoning, text-to-speech, turn detection and interruption handling in one connection, at $0.075/minute
- Speech Understanding add-ons: speaker diarization, sentiment analysis, auto chapters, topic detection, summarization and entity detection, each priced per hour on top of the base transcription rate
- Guardrails for PII audio/text redaction, profanity filtering and content moderation
- LLM Gateway giving direct, billed-per-token access to OpenAI, Anthropic, Google and other model providers alongside transcripts
Pricing
- Free: $50 in free credit on signup, no credit card required, capped at 5 new streams/minute
- Pre-recorded transcription: from $0.15/hour (Universal-2) to $0.21/hour (Universal-3.5 Pro)
- Streaming transcription: from $0.15/hour (Universal-Streaming) to $0.45/hour base (Universal-3.5 Pro Realtime)
- Voice Agent API: $0.075/minute ($4.50/hour) all-inclusive
- Add-ons such as diarization, redaction, translation and sentiment stack on top of the base rate, typically $0.01-$0.15/hour
- There is no flat monthly plan; billing is entirely pay-as-you-go and scales with volume, so check AssemblyAI's pricing page for current per-hour rates
Who it's for
The $50 free credit is enough for developers to prototype a transcription or voice-agent feature before entering a card. Pay-as-you-go pricing suits startups and scale-ups building voice products, such as call centers, meeting tools, or media captioning, who want to pay for what they process rather than commit to a seat-based SaaS plan. Teams needing compliance features like PII redaction or content moderation, or wanting to pair different LLMs with the same voice pipeline, are the target audience for the add-on and LLM Gateway layers.