Plans and tiers
You can run inference on our shared compute: you send a request to the API, and it is served on hardware we run and keep loaded. That access is sold in four subscription tiers — Bronze, Silver, Gold, and Platinum.
When you sign up you enter the free Bronze tier. It takes no card. You get a base URL and an API key, and you can start making requests within Bronze's limits.
What the tiers differ on
Six things change as you move up:
- Concurrency — how many requests you can have in flight at once.
- Request rate — how many you can send a minute.
- Context window — how much context one request can carry.
- Included tokens — how many you can spend a month.
- Routing priority — whether your requests go first when the pool is busy.
- Service commitments — a rate-limit pool of your own, an SLA, and a named support contact.
See Compare tiers for what each tier includes.
Silver and Gold are bought in the product and billed monthly. Platinum is quoted per account.
Every limit is enforced. When you reach one, the request is refused rather than slowed, and the error names the limit you hit. See Errors.
Tiers buy throughput, not models
Moving up a tier does not usually add models. Two neighbouring tiers can serve exactly the same ones and differ only in the limits above.
To find out whether you can call a particular model, check the model catalog, which lists the tier each model needs. To find out how much you can call it, check the compare page.
What your plan does not control
Dedicated hardware. Running a model on a node of your own is a separate purchase, not something a tier unlocks. You can be on Bronze and own one. See Shared and dedicated.
Your own model weights. No shared tier runs weights you supply. Dedicated hardware does, including a model we do not carry in the catalog — open a support ticket to arrange it.
Changing tier
Upgrade or downgrade under billing in the product. The change takes effect when the payment goes through, not at the end of the billing period.
Your API limits change almost immediately. The product interface can take a few minutes to catch up, because your session holds a copy of your plan until it refreshes; signing out and back in updates it.