Use this whenever a pipeline depends on a specific AI model name hardcoded into a request, which is nearly all of them.
Steps
- When an API call starts returning a quota error, check whether the limit is a normal rate limit (temporary, resets soon) or a hard 0 limit (the model itself has lost free-tier access, waiting won't help).
- A limit of exactly 0 for a specific model, especially paired with a request to "check your plan," usually means the model was deprecated and quota was pulled, not that you're being rate-limited.
- Check the provider's current model list and pick the current recommended equivalent (often a very similar name, like a version bump).
- Swap the model name in the request, retest immediately, since prompts and expected response shapes are often close enough to work unchanged.
Gotcha
It's tempting to assume a working integration is done and won't need revisiting. Model deprecation happens on the provider's timeline, not yours, sometimes with only a few weeks of public notice before free-tier quota actually drops. Treat the specific model name in any request as a value to periodically verify, not permanent configuration.
Quick reference
Symptom: limit: 0 on a specific model name, not a temporary rate-limit countdown.
Fix: check for a deprecation notice, swap to the current model, retest.