Groq's model lineup rotated — the current chat models are gpt-oss-120b, gpt-oss-20b and qwen3.6-27b, and the Llama-era IDs that every LiteLLM tutorial hardcodes are no longer in the catalog. You can check what your key actually has with one call:
curl -s https://api.groq.com/openai/v1/models -H "Authorization: Bearer $GROQ_API_KEY"
If you call the old ID directly, you at least get a loud error. The sneaky case is a LiteLLM fallback chain. The whole point of fallbacks is to absorb failures — so when the first entry in the chain is a dead model ID, every single request fails over to your second choice, silently. Nothing crashes. Your app keeps answering. You just quietly lose the provider you picked for speed, eat an extra failed round-trip of latency on every call, and shift 100% of traffic onto the fallback (and its rate limits) without a word in your logs unless you look for it.
Worth two minutes today: grep your configs for llama-3.3-70b-versatile and mixtral-8x7b-32768, swap in a model ID from the live catalog, and if you run fallback chains, alert on fallback-hit rate — it is the only place this class of rot shows up.
How the LiteLLM free-tier setup fits together: https://toolfreebie.com/litellm-free-llm-gateway/
Top comments (0)