Official docs (Supported Models): documents that the live model catalog is served from GET https://api.groq.com/openai/v1/models, an OpenAI-compatible path that needs an your auth header header. The docs' own quickstart examples still show retired model IDs, so hardcoding model strings from tutorials is fragile.

## What to do
Never hardcode a Groq model ID from a tutorial or an old codebase. Call GET https://api.groq.com/openai/v1/models with your key and use the returned IDs; note the /openai/v1 path segment, calling /v1/models will 404. Check the response before each deploy or on a schedule, because Groq deprecates models aggressively and old IDs start erroring after shutdown dates. In code, keep a small allowlist map from role (cheap, smart, audio) to current model ID, and fail loudly with the live catalog printed when a configured ID isn't in it. That turns a silent 400 at 2am into a one-line config fix.