Your LLM Fallback Probably Isn't a Fallback

Short summary

A post-mortem of an outage caused when DeepSeek retired two API model names (deepseek-chat and deepseek-reasoner) and replaced them with deepseek-v4-pro and deepseek-v4-flash. The configured fallback failed because it shared the same provider and API key, carrying the identical 400 error. The fix was repointing gateway aliases, and the author provides a concrete checklist for validating your own LLM fallback setup.

• •DeepSeek retired deepseek-chat and deepseek-reasoner API names, breaking a live LLM gateway
• •Fallback failed because it shared the same provider and API key as the primary
• •Author provides a checklist: cross-provider fallback, reachability canaries, and CI gate detection

Generated with AI, which can make mistakes.