What
When Claude reads a whole PDF, Claude Code used to have a fallback that turned the PDF's pages into images on your machine first (page extraction). That fallback and the tengu_deep_starlight check guarding it are removed.
- Whole-PDF reads now go straight to the check of whether the model supports reading PDFs, and then to the model's own PDF reading.
- The page images are never produced: the render setting is always "none" and no pages are counted as rendered.
- The fallback only ran when the
tengu_deep_starlightcheck came back false, and its default was true, so it ran only where a server setting had turned it to false. Nothing has been read about howtengu_deep_starlightortengu_pdf_page_extractionwere set.
Why
For most people this changes little, since the fallback was only reachable through a server setting. Where it was in use, PDFs that are too large or on models that cannot read whole PDFs will now get the message 'Reading full PDFs is not supported with this model.' instead of being converted to images.
tengu_deep_starlight 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 v2.1.307: 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 v2.1.307. It isn't a statement about your account. What a flag value here can and cannot tell you