Nothing to try yetTier: how much it should matter to you
3Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
CompactionArea: what it touches
ImprovementsKind: in v2.1.283,
ImprovementsSection of the release
What
Auto-compact summarizes older parts of a conversation when it grows too long for the model's context window, which is how much text the model can take in at once. A cap on that window, which previously could never apply, can now be set from the server. It is only considered when all of these hold:
the model natively supports a context window of at least 1M tokens
A value of "600k" or "500k" limits the window to 600,000 or 500,000 tokens. The built-in default is empty, so no cap applies unless the server sends a value.
Why
Desktop users on 1M-context models could see compaction start earlier if the server applies the cap to their account.
Read from
Feature flag
tengu_goofy_jellyfishNot enough to say
Nothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.
This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.283: not a boolean we can read
These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.
