What
When a model refuses a request, Claude Code can switch to a fallback model. The effort level, which controls how much reasoning the model puts in, can now carry over to that fallback model.
- The request now includes
refusalFallbackEffort, and the switch records the effort that was carried over. - The saved fallback choice now stores
carriedEffort. - The effort is not carried over when the reason for the switch is
hook,turn_layer,none_sentorkilled. - New helpers can undo the switch to the fallback model when you pick a model yourself (
unlatchRefusalFallbackModel, reason"model_pick"), and can clear the carried effort. - Claude Code now records the carried effort level after a successful request, along with whether it was used and whether it has ended.
This code refers to a feature gate, a switch Anthropic can turn on or off remotely. Its state has not been read and is unknown.
Why
If this applies to you, a fallback model should keep the reasoning effort you set instead of using a different one.
tengu_loggia_passage Not enough to sayNothing 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 v2.1.287: off
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.
Read once, for one account on one subscription tier, against v2.1.287. It isn't a statement about your account. What a flag value here can and cannot tell you
It is not clear whether the carry-over, or the refusal fallback itself, is switched on for users.
New in this build: tengu_loggia_passage