Unclear What makes system/init report ultrathink_typed_raises as true is not shown, so it is unclear when a host can use this.
The SDK lets other programs, called hosts, build on Claude Code. A host can now send an optional ultrathink_typed value with a user message. It tells Claude Code that the person really typed the word "ultrathink" and saw it highlighted in colour.
Claude Code acts on it only when the message that opens the session, system/init, included ultrathink_typed_raises: true. When it does, the turn runs at xhigh effort, meaning the model spends more reasoning on it, as long as:
- Claude Code finds the word in the message text
- the text was not pasted
- nothing else suggests the text was not typed
Hosts can now let a typed "ultrathink" raise effort to xhigh. The checks on pasted or untyped text keep it from raising effort when the word was not actually typed.
What makes system/init report ultrathink_typed_raises as true is not shown, so it is unclear when a host can use this.