SILENT DEPRECATIONS BREAK PRODUCTION

Which AI models is your production actually pinned to?

Model vendors deprecate, reprice, and silently update models on their own schedule — and your app finds out when it breaks at 2am. Run this 30-second audit in your repo to list every model your code depends on. No signup, nothing installed.

The 30-second dependency audit

1

Find hardcoded model names in your repo:

grep -rhoE "gpt-[0-9][a-z0-9.-]*|claude-[a-z0-9.-]+|gemini-[a-z0-9.-]+|llama-[a-z0-9.-]+" . --exclude-dir=.git --exclude-dir=node_modules 2>/dev/null | sort | uniq -c | sort -rn | head -20
Names with no version pin (e.g. "gpt-4" with no date) → RISK. Unpinned aliases get silently swapped under you.
2

Check what your environment actually points at (values masked — safe to run):

env | grep -iE "model|OPENAI|ANTHROPIC|GEMINI" | sed -E 's/=(.{4}).*/=\1***/' | head -20
Model set only in one dev's shell, not in config → RISK. If it's not in versioned config, it doesn't exist when it matters.
3

Find SDKs that auto-upgrade models:

grep -rnE "\"openai\"|'openai'|\"anthropic\"|@google/generative-ai" package.json requirements.txt pyproject.toml 2>/dev/null | head; grep -rn "latest\|\"*\"" package.json 2>/dev/null | grep -iE "openai|anthropic" | head -5
Floating SDK versions + unpinned models → double exposure. An SDK bump and a model swap can land in the same deploy.
Every model pinned, versioned, and documented → you're ahead of most teams. Now make sure someone watches the vendors' changelogs.

This audit shows what you're pinned to today. It can't tell you what the vendor changes tomorrow — that's the part that needs watching.

If you found unpinned models, do this now

Watch your models free → ModelMinder

This page audited what you're pinned to. ModelMinder watches the vendors for you — deprecations, repricing, and silent updates — and alerts you before production breaks.

Share this check: XLinkedInReddit