Your first US$10 is on us: a new organisation gets US$10 of credit when OS3 sets it up. After that, add a card or buy credit in Usage. A call with no credit left and no card is refused with
402 payment_required.Set up
Create a key in the portal (Keys → Create key) and keep it in a.env file, out of version control:
.env
Send a chat completion
Your key is the API key; the model is its id.x-os3-request-id header. Send it to support when a request goes wrong.
Let it think
The model thinks unless you sendreasoning_effort: "none". Efforts are low, high (the default) and max. Thinking comes back in reasoning_content, and its tokens are billed as output. Stream long answers.
Anthropic’s Messages API
The same model answers in Anthropic’s shape. Use the gateway’s address without/v1. Thinking comes back as a thinking block with an empty signature.
OpenAI’s Responses API
OS3 keeps no response: send the conversation each turn. A request withstore: true or previous_response_id is refused.
The fast id
The same model on a separate, less crowded lane. Each answer is billed by the lane that served it, and thex-os3-lane header says which. See Normal and fast.
Next
- Pricing and limits by plan.
- Errors, with what to do for each.
GET /v1/modelslists the ids your key can call.