The BastionGPT API uses simple, usage-based pricing: a base of $45 per month that covers roughly your first 2 million words processed each month, then about $20 per million words beyond that. "Words processed" counts your input and the AI's output together, and across all of our customers the average request works out to about two cents.
Here is how each part of BastionGPT API pricing works, what lowers your bill, and what we never charge you for.
How is the BastionGPT API priced?
You will not find a per-model, per-token price list, and that is deliberate. Our orchestration layer automatically routes each request to the best AI model for the task (see What AI models power BastionGPT?), so you pay one predictable rate no matter which model does the work:
Base: $45 per month, covering roughly 2 million words processed
Above the base: about $20 per million words, input plus output combined
In practice: a short question costs a small fraction of a cent, and a request carrying a long clinical history costs a few cents
That rate is the full cost of a HIPAA-compliant AI API. Registering for API access incorporates the same HIPAA agreements that cover the rest of BastionGPT, the Business Associate Agreement is included, and there is no separate compliance fee.
Does prompt caching lower my bill?
Yes. The API is stateless, so you resend your instructions and context with every call, but resending does not mean repaying. Prompt caching is automatic: when a request contains content our system has recently processed, you are not charged for that cached portion. To benefit as much as possible, keep the parts of your request that never change (instructions, templates, reference documents) at the front, and put anything that varies, such as today's date or the actual question, at the end.
How much does audio transcription cost through the API?
The same engine behind our AI Scribe is available through the API, so your app can send audio for HIPAA-compliant transcription and clinical note generation. Audio costs about $0.73 per hour, billed per minute, so a five-minute recording is billed as five minutes rather than a full hour.
What are you never charged for?
Our system sometimes reprocesses a request behind the scenes. If our quality checks detect signs of an unreliable response, we discard it and rerun your request on another model. You are never billed for that internal reprocessing: your charge reflects your initial request and the final output, nothing in between.
How do statements and usage reports work?
Monthly statement. Within the first week of each month, we send a consumption statement covering the prior month's usage.
Per-key reporting. Usage is tracked separately for each API key, and our support team can prepare an ad hoc readout for any key, including a breakdown of how much of your usage was cached.
Usage audits. After a month or two of steady use, we can audit your workload. We never compromise on quality to save money, but if your requests consistently run on lower-cost models, we can apply a discount factor to your rate.
Is the API billed separately from my BastionGPT subscription?
No, it is all one account. If you already subscribe to BastionGPT, enabling API access moves your plan to the usage-based model above, and your web app use and API use are combined on a single monthly statement. Most single-provider accounts never exceed the included volume, so their bill simply stays at the monthly base.
Are there rate limits?
Yes, one per API key, and they exist to protect you rather than to limit you. If a bug in your code starts looping requests, the rate limit turns a potential surprise bill into a clear notice from the API. If your legitimate usage grows into a limit, contact us and we can raise it for the specific keys that need it. Every account starts with two identical keys, commonly used as test and production, and we can attach up to 20 keys if you manage multiple apps or clients.
How can I test the API before standard billing begins?
New API accounts include a 30-day developer trial with $20 in free tokens, enough to build and test a working integration. Access starts with a short call about your use case, and once you are approved, your keys usually arrive within about one business day, along with Swagger documentation and a Postman collection. When you are ready, you can apply for API access on our website. You can read more in Is there an API? I want to use BastionGPT in my app. And because the same API powers the BastionGPT web app, starting a free trial of the app gives you a no-code way to preview exactly the response quality your integration will deliver.
If you would like help estimating costs for your specific workload, email [email protected] and we will walk through it with you.