Update _models.py to handle Python 3.14

[Natim]

• I understand that this repository is auto-generated and my pull request may not be merged

Changes being requested

Please using Annotated[Union] with Python 3.14

Additional context & links

Fixed #2704 to work properly with Python 3.14

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Restore caching of union discriminator metadata

The new code replaces cast(CachedDiscriminatorType, union).__discriminator__ = details with a bare cast(CachedDiscriminatorType, Annotated[union, details]). This only constructs an Annotated type and discards it, so the union instance never receives a __discriminator__ attribute. Subsequent calls to _build_discriminated_union_meta will therefore never hit the isinstance(union, CachedDiscriminatorType) fast path and tests that expect UnionType.__discriminator__ to exist (e.g. tests/test_models.py::test_discriminated_unions_invalid_data_uses_cache) will fail. Either the attribute assignment needs to be restored or the rest of the code must be updated to read metadata from the Annotated type instead.

Useful? React with 👍 / 👎.

[1155159410]

Thanks for this PR. I think it’s quite urgent since the current OpenAI API isn’t working on Python 3.14 because of this issue.

[RobertCraigie]

Thanks for the PR but this isn't the correct fix, the places that read the __discriminator__ property need to be updated as well

[Natim]

The discriminator property need to be updated as well

Would you like for me to do so?
My understanding was that some of the code was generated automatically and PR couldn't be approved directly like that, reason why I didn't put much effort fixing everything related to it.