Pricing built for innovation, not contract negotiation

Universal-3 Pro is our most advanced speech language model, designed specifically for speech tasks. It uses a prompt-based architecture for deeper contextual understanding and allows domain-specific customization—no retraining needed. Universal-2 is a high-accuracy model supporting 99 languages, built for general-purpose use cases. It offers strong out-of-the-box performance and supports features like speaker diarization and real-time streaming. Perfect for legal, medical, and other specialized use cases. Universal-Streaming is an ultra-fast, ultra-accurate streaming speech-to-text model designed for voice agents.

Yes! With the free offer, you get $50 in credits to use towards AssemblyAI’s Speech-to-Text APIs. To add more credits, simply add a credit card to your account.

Absolutely! If you plan to send large volumes of audio and video content through our API, please reach out to us here to see if you qualify for a volume discount.

We don't limit how many streams you can run simultaneously - only how quickly you can start new ones, giving you unlimited scale while ensuring reliable performance.
‍
Free users can start 5 new streams per minute, while pay-as-you-go accounts start with 100 new streams per minute. Anytime you are using 70% or more of your current limit, your new sessions rate limit will automatically increase and scale up by 10% every 60 seconds. This means within 5 minutes of sustained usage, you can scale from 100 to 146 new streams per minute (for a total of 610 concurrent streams), with unlimited ceiling as your usage grows.

These limits are designed to never interfere with legitimate applications - normal scaling patterns automatically get more capacity before hitting any walls, while protecting against runaway scripts or abuse. Your baseline limit is guaranteed and never decreases, so you can scale smoothly from dozens to thousands of simultaneous streams without artificial barriers or surprise fees.

Need higher limits? Contact our sales team for custom limits that match your deployment timeline.

We charge based on total session duration - the entire time your connection stays open, whether audio is flowing or not. This gives you complete transparency and control: you pay for exactly what you're using, with no hidden costs for idle streams. You can choose to keep streams open continuously for instant response or open them strategically as needed to minimize costs, scaling up and down without prepaid commitments based on how your voice application actually works.

Most audio files sent to AssemblyAI's API can be processed in less than 60 seconds. For example, you ca process a 30 minute pre-recorded audio file in 23 seconds with Universal speech-to-text model.

Great question. Once you add a credit card and deposit funds into your account, your account's funds will be drained as you use the API.

When multichannel is enabled, each channel will be transcribed and billed separately. The total cost is calculated by taking the hourly transcription rate (billed per second) and multiplying it by the number of channels. To calculate your total cost, simply multiply your recording's duration by the hourly rate, then multiply that result by the number of channels.

For example, if you sent a 5-minute recording with three channels, you would be billed for the 5 minutes of audio multiplied by the standard rate, with that total multiplied by three channels. This is equivalent to being billed for 15 minutes of audio.

You can also get started with AssemblyAI on the AWS Marketplace—or ask your AWS account team about how to leverage AssemblyAI to revolutionize the way your company understands its customers.

Feel free to email us at support@assemblyai.com, or click the chat button in the bottom right corner of your browser to chat live with our API Support team!

We support over 99 languages and counting, including Global English (English and all of its accents).

In the context of a Large Language Model (LLM), a “token” is the smallest unit of text processed by the model. 100 tokens roughly maps to ~75 words.