Pricing
Pay per token. Cache hits at 10%.
Every model is priced per million tokens. Input that repeats a recent prefix is billed at the cached rate automatically. There are no minimums, seat fees or platform charges, and policy evaluation is included.
Estimate your bill
- Per request
- $0.004
- Requests per month
- 150,000
- Without caching
- $765
- Saved by caching
- $194
Placeholder rates. No minimums, no seat fees; you pay for tokens the model server reports.
Questions
How are tokens counted?
From the usage the model server reports on each response, the same numbers you see in usage. Prompt tokens that hit the server's prefix cache are reported as cached_tokens and billed at the cached rate.
Is policy evaluation extra?
No. Built-in PII redaction, the platform baseline, and the per-request decision headers are included. If you run your own external classifiers, you host those.
Do blocked requests cost anything?
A request blocked at the input stage never reaches a model and costs nothing. A response blocked at the output stage was generated, so its tokens are billed like any other.
What about rate limits and concurrency?
Defaults are 600 requests per minute and 16 concurrent requests per key, adjustable per key. Higher ceilings for sustained volume are a conversation, not a tier.
Are these the final prices?
Not yet. The figures on this page are placeholders while we finalise capacity. Keys you create now keep working when final pricing is published.
Ready? Create a key and your first requests show up on the dashboard within a minute.