Failed-request billing

You are billed for work the provider performed, not for the status code.

Situation	Billed?
Request rejected before inference, validation error (`400`), auth failure (`401`/`403`), unknown route (`404`)	❌ No
Rate limited (`429`) or service unavailable (`503`) before the model ran	❌ No
Out of balance or allowance (`429 insufficient_quota`), the request never runs	❌ No
Model started generating then the stream failed partway	⚠️ Partial, you are billed for tokens actually produced
Successful response (`200`)	✅ Yes

Streams

Tokens generated before a stream fails are real work and are counted in usage. Read the usage block on the final chunk or the error payload rather than assuming a failed request cost nothing. If a stream ends before the provider reports usage, the gateway settles from what it can prove and never bills work that did not dispatch. Reconcile against the developer console and agent status, not status codes.

Get balance
Get usage
Error format

Basics

Features

Server-Side Tools

Multimodal

Best Practices

SDKs

API Reference

Failed-request billing

Streams

​Streams

​Related

Streams

Related