$ the-wire · showcase
openai-node 7.23.0 adds Cyber access programs and a GPT-Rosalind model
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
![openai-sdks[bot]](https://avatars.githubusercontent.com/in/3705508?v=4)
10 people shipped this
copyberry
@copyberry
1 cited
Justin Beckwith
@jbeckwith-oai
1 cited
openai-sdks[bot]
@openai-sdks[bot]
1 cited
HughChaw
@Hughhhhcoder
1 cited
Minh Vu
@fallintoplace
1 cited
johnl-oai
@johnl-oai
Ahmed Ibrahim
@aibrahim-oai
Ruslan Nigmatullin
@euroelessar
marcuswood-oai
@marcuswood-oai
markstuart-oai
@markstuart-oai
The Responses API gained typed Cyber access program selection in both SDKs while the Python client fixed header merging and single-pass tool iterables, and Codex tightened sandbox credential and Guardian review handling.
v7.23.0 openai/openai-node
Version 7.23.0 ships typed support for GCP external storage and a GPT-Rosalind research model, alongside the Cyber access program fields. Node users get the whole set by bumping to 7.23.0.
fix(chat): preserve single-pass tool iterables (#3770) openai/openai-python
Passing a generator to chat.completions.parse() or .stream() could send an empty tools array or drop parsed function arguments, because the single-pass iterable was consumed more than once. The fix materializes it once and reuses it across validation, serialization, and response parsing; async .stream() now also accepts non-strict tools without auto-parsing them, matching sync behavior.
fix(client): merge HTTP headers case-insensitively (#3486) openai/openai-python
HTTP header names are case-insensitive, but differently cased overrides left duplicate values behind and Omit() could fail to remove an earlier one. Headers now merge case-insensitively so later values replace earlier ones and explicit removal respects casing, covered by a parametrized regression test for sync and async clients.
feat(api): add Cyber access programs to Responses openai/openai-node
Response creation options accept access_programs.cyber, the resolved program is exposed on response resources, and supported values are typed as standard, daybreak_blue, and daybreak_red. openai-python carries the same fields.
Repair rejected Windows sandbox credentials during provisioning openai/codex
Windows sandbox setup could register as complete even when Windows had rejected the stored account passwords, since account flags alone missed the failure before runtime logon. Provisioning now checks stored credentials for both offline and online accounts and classifies logon failure, expired-password, and password-change-required errors. Also worth noting: Codex extended Guardian computer-use...
