{"version":"2.1.307","anchor":"auto-compact-window-persisted-per-model-via-internal-metadat","canonical_anchor":"auto-compact-window-persisted-per-model-via-internal-metadat","heading":"Auto-compact window is remembered per model when a session is resumed","tier":"notice","area":"Compaction","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.307\/e\/auto-compact-window-persisted-per-model-via-internal-metadat","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.307","markdown":"### Auto-compact window is remembered per model when a session is resumed\n\nEach model's auto-compact window is now saved with the session and restored on resume, and the default window for a model has changed\n\n**Unclear.** The size limits and cap behind the new default window are not known, so it is unclear when compaction now starts for a given model.\n\n**What**\n\nCompaction is when Claude Code summarises older parts of a conversation to free up room in the context window, which is how much text the model can take in at once. The `autoCompactWindow` setting decides how full the context gets before that happens.\n\nClaude Code now saves a value for each model with the session's own stored data. The value is either a number or \"auto\". When the session is loaded again, that saved value is applied as if you had set `autoCompactWindow` in your user settings. Changes to it are written back to the session.\n\nThe default window for a model also works differently now:\n\n- A model whose window is above a certain size is given a cap.\n\n- Other models default to a 1M window that you opt into.\n\n**Why**\n\nA compaction window chosen for a model carries over when you resume a session, instead of being lost.\n\n- Area: Compaction\n- Names: `autoCompactWindow`\n- Tier: You'll notice\n- Useful: 3\/5\n- Signal: 1\/5\n- Scope: individual\n- Heads-up: no"}