You'll noticeTier: how much it should matter to you
2Useful: my rating, 1 to 5
2Signal: worth watching, 1 to 5
ModelsArea: what it touches
ImprovementsKind: in v2.1.285,
ImprovementsSection of the release
What
When you use Claude through AWS or Google Cloud credentials, Claude Code checks at start-up which models your account can use. It now saves each result, allowed or refused, per account in a cache/model-access.json file and can reuse it on later starts.
To tell accounts apart, Claude Code builds a fingerprint of your AWS config and credentials files or your gcloud identity files. If one of those files is over 4 MB or is not a regular file, saved results are not reused.
Start-up can skip some model checks, so it may be quicker. A saved answer can go out of date, though: if your model access changes and Claude Code still behaves as before, set CLAUDE_CODE_SKIP_MODEL_ACCESS_MEMORY.
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtIt is not clear how long saved results are kept or exactly which providers are covered.
