feat(codex): add gpt-image-2.5, flare, and sunburst image models support

[yheiakadylan]

No description provided.

[yheiakadylan]

Summary

This PR adds official support for OpenAI Codex's latest image generation models (gpt-image-2.5, gpt-image-2.5-flare, and gpt-image-2.5-sunburst), as well as complementary image tool models (gpt-image-2, gpt-image-1.5).

It also updates the Codex image provider handler to properly route tool-based image models through ChatGPT's Responses API while maintaining full backward compatibility with legacy -image suffixed models.

---

Background & Problem

OpenAI rolled out new image generation models for ChatGPT/Codex subscribers under the gpt-image-2.5 family (gpt-image-2.5, gpt-image-2.5-flare, and gpt-image-2.5-sunburst).

Unlike legacy models which simply strip the -image suffix to target a chat model (e.g., gpt-5.5-image → gpt-5.5), tool-based image models:

1. Require the top-level Responses API model parameter to be set to a supported host model (gpt-5.5).
2. Require the image_generation tool definition to explicitly pass the target model (imgTool.model = toolModel) and action (generate or edit depending on image inputs).
3. Require explicit tool_choice and reasoning settings to ensure the Responses API triggers the image generation tool properly.

---

Changes

1. Codex Image Provider Handler (open-sse/handlers/imageProviders/codex.js)

• Introduced CODEX_TOOL_IMAGE_MODELS set containing gpt-image-1.5, gpt-image-2, gpt-image-2.5, gpt-image-2.5-flare, and gpt-image-2.5-sunburst.
• Implemented resolveCodexImageModels() to differentiate between:
  • Tool models: mapped to host gpt-5.5 with imgTool.model = toolModel and imgTool.action = (refs.length > 0 ? "edit" : "generate").
  • Suffix models: cleanly stripped (e.g., gpt-5.5-image → gpt-5.5) without breaking existing workflows.
• Configured tool_choice: { type: "image_generation" } and reasoning effort for tool models so prompts consistently trigger image generation.

2. Provider Registries

• open-sse/providers/registry/codex.js:
  • Registered gpt-image-2.5, gpt-image-2.5-flare, gpt-image-2.5-sunburst, gpt-image-2, and gpt-image-1.5 with capabilities: ["text2img", "edit", "multiImage"].
• open-sse/providers/registry/openai.js:
  • Added corresponding gpt-image-2.5, gpt-image-2.5-flare, and gpt-image-2.5-sunburst entries under the OpenAI provider registry for consistency.

3. Unit Tests (tests/unit/image-generation.test.js)

• Added comprehensive unit test covering Codex image generation using gpt-image-2.5 tool model to verify:
  • Responses API endpoint and headers.
  • Host model resolution (gpt-5.5).
  • Correct tool payload ({ type: "image_generation", model: "gpt-image-2.5", action: "generate" }).
  • Correct tool_choice and reasoning configuration.

---

Verification & Testing

• Syntax validation checked via node --check.
• Unit test passing for gpt-image-2.5 tool call payload formatting.
• Verified that existing models (gpt-5.5-image, gpt-5.6-sol-image, etc.) remain unaffected.
• Clean commit history branching directly off upstream/master without extraneous repository modifications.

[decolua]

Thanks @yheiakadylan for the contribution! Reviewed and merged into master. 🙏