[ win Codex ] Expose effective image model and supported model selection for built-in image_gen (Images 2.5)

Request

Please expose the actual image model used by Codex Desktop's built-in image generation tool, and a supported model selector where the account has access to multiple image models.

Environment and observed interface

• Windows x64
• Codex Desktop 26.901.6511.0
• Bundled Codex backend 0.153.4
• Inspected on 2026-09-09

The caller-visible built-in image_gen.imagegen schema provides only:

prompt
referenced_image_paths
num_last_images_to_include

There is no caller-visible model field. Direct Images API parameters cannot be passed through this schema.

The Images 2.5 announcement says the rollout includes Codex:
https://openai.com/index/introducing-chatgpt-images-2-5/

The Codex image-generation guide inspected on the same date still describes the built-in model as gpt-image-2:
https://learn.chatgpt.com/docs/image-generation

Reproduction

1. Open a Codex Desktop task with the built-in image-generation capability.
2. Inspect its available input schema.
3. Try to determine or explicitly select which Images 2.5 model will handle a generation/edit before calling it.
4. Observe that the input schema has no model selection or capability query.

Expected behavior

• Provide supported model availability/selection, or clearly document that the built-in route is automatically managed.
• Expose safe effective-model metadata for a generation/edit so the user can verify what ran.
• Keep the built-in tool documentation and rollout/capability information consistent.
• If exact model selection is intentionally API-only, explicitly state this in the built-in tool documentation.

Evidence limits

This is an interface/observability request, not a claim that the backend is using an old model or that generation is broken. No image-generation request was made for this inspection; result metadata was not tested. The missing input selector alone cannot establish the actual backend model or account rollout status.

Submitted by the user's Codex assistant with explicit authorization. No private task contents, local paths, account identifiers, or credentials are included.

Start by reviewing the built-in image_gen.imagegen input schema and the linked image-generation guide, then compare the requested model selection and effective-model metadata with the existing Codex Desktop interface. Done means supported selection or capability information is exposed safely, effective-model reporting is available, and the tool documentation matches the rollout behavior; validate with a Codex Desktop task without assuming backend behavior.