When Claude Code runs non-interactively, it can report running totals while the model is still working on a reply: cost so far in US dollars, tokens used and cache use. This applies to print mode and to programs built on the SDK, Anthropic's toolkit for driving Claude Code from code.
These mid-turn updates used to go out at most every 15 seconds for everyone. Now how often they run comes from a value the server sends.
- No value, zero, or anything that is not a positive number: mid-turn updates are skipped entirely.
- Any positive value: updates run, but never more often than every 15 seconds.
The built-in fallback is zero, so these updates are off unless they are switched on remotely. Totals can still be sent by the other path that reports changes.
Anything that showed live cost or token counts while a reply was in progress may stop seeing those numbers update until later. Watch for this if you track spending in real time in a remote or SDK session.
tengu_enchanted_willow 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 v2.1.291: 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.
Read once, for one account on one subscription tier, against v2.1.291. It isn't a statement about your account. What a flag value here can and cannot tell you
New in this build: tengu_enchanted_willow