Desktop App
Max mode
Max mode opts into full context window usage and disables automatic context compression safeguards.
What max mode changes
- Uses the model full context window instead of the default constrained effective window.
- Disables auto-compaction for requests while max mode is enabled.
- Disables other context-compression behavior intended to keep context lean by default.
Default mode vs max mode
- Default behavior applies a smaller effective context target (currently min of 80k tokens or 60% of model context).
- Max mode removes that reduction and uses full available context length.
- Larger context often improves continuity, but increases token usage and cost per turn.
Availability notes
- The max mode toggle is shown for non-Claude Code model selections in current UI.
- Model/mode combinations may differ across local CLI harness paths, so validate behavior in your chosen model.
When to use max mode
- Use for complex sessions where long continuity is more important than token efficiency.
- Avoid for routine short tasks where default context shaping is sufficient.
- Turn it off again after high-context tasks to reduce ongoing spend.