Model and API deprecation policy
How we retire models, change endpoints and version the API, and what happens to your requests when that occurs.
Advance notice
We publish a notice at least 30 days before a model is retired or an endpoint changes. The notice appears on /status and in the Changelog/Notices section of /llms.txt.
Until the retirement date the model stays on /pricing with a note that it is being withdrawn.
Requests after retirement
After retirement, requests to the model return an error with a clear code and a link to /pricing for alternatives.
Your unused balance does not expire and is not affected by a model retirement.
Deprecation headers
Where the upstream provider sends Sunset or Deprecation headers, the gateway passes them through. The gateway does not control upstream headers, so for all other cases the retirement date is published on /status.
Rate limits and response headers
Rate limits apply per API key and are counted in two dimensions: requests per minute and tokens per minute. The limit depends on your lifetime spend tier: Tier 0 (at registration) is 200 requests per minute; $100 spent gives 500; $500 gives 2,000; $2,500 gives 5,000; $10,000 gives 10,000; $25,000 gives 20,000. Tiers rise automatically and are not lowered.
Inference responses include x-ratelimit-limit-requests, x-ratelimit-remaining-requests and x-ratelimit-reset-requests. Read the remaining values on every response and slow down before the limit is reached. Model-list requests (GET /v1/models) do not return these headers.
When the limit is exceeded, the gateway returns HTTP 429 with the body {"error":{"type":"rate_limit_error","code":"rate_limit_exceeded"}} and a Retry-After header in seconds. The request is not processed, so a retry is safe.
Recommended client behaviour: on 429, wait for Retry-After if it is present, otherwise use exponential backoff with jitter; stop after 3 to 5 retries. Retry only on 429, 5xx and timeouts.
Browser clients: requests from browser pages are accepted only from our own site origins (access-control-allow-origin header). Browser apps on other domains need their own server-side proxy. Server-side integrations (SDKs, bots, backends) have no such restriction.
API versioning
The current API lives under the /v1 path. A major version change creates a new base path (for example /v2).
The previous major version remains available for at least 6 months after the change is announced.
Questions
Write to support@llmka.dev with the model identifier if you need help migrating. See /docs and /pricing for current models.