Add OpenAI Responses API support with chat_completions capability flag

[mstallone]

Add support for OpenAI's /responses endpoint for models that don't support /chat/completions API. This enables compatibility with newer model variants (gpt-5-codex, gpt-5-pro, o3-pro, etc) while maintaining compatibility with existing configs

Changes:

• Add supports_chat_completions flag to model capabilities that defaults to true for existing behavior
• Implement responses API client with streaming support as per OpenAI documentation.
• Add ResponseEventMapper to convert responses events to completion events for maintainer simplicity
• Update UI to allow toggling chat_completions capability
• Add gpt-5-codex model

Closes #38858

Release Notes:

• Added support for OpenAI's /responses endpoint for models that don't support /chat/completions API.
• Added OpenAI gpt-5-codex model to assistant

[somtooo]

@mstallone Quick Question: I built your branch but still get the error

Failed to connect to API: 400 Bad Request {"error":{"message":"model gpt-5-codex is not accessible via the /chat/completions endpoint","code":"unsupported_api_for_model"}}

Dont think this pr solves the copilot chat problem. Looks like it adds support for OpenAi/ openAi compatible providers hence it does not close #38858 , but nice to have the code that supports /responses. Thanks. Would be less difficult to extend to copilot chat

[mstallone]

Hi @somtooo, thanks for giving the branch a test! I am not sure I understand what you mean about the copilot chat.

I added some screenshots below showing that the chat does work with attachments on a fresh install from the branch:

OpenAI as a provider with gpt-5-codex builtin:

Azure OpenAI as a third-party OpenAI compatible endpoint with a custom gpt-5-codex deployment:

For OpenAI compatible, you must set chat_completions to false otherwise it will try the completions endpoint.

"openai_compatible": {
  "Azure OpenAI": {
    "api_url": "https://zed.openai.azure.com/openai/v1/",
      "available_models": [
        {
          "name": "gpt-5-codex",
          "display_name": "gpt-5-codex - Azure OpenAI",
          "max_tokens": 272000,
          "max_output_tokens": 128000,
          "max_completion_tokens": 128000,
          "capabilities": {
            "tools": true,
            "images": true,
            "chat_completions": false
        }
      }
    ]
  }
}

If you could let me know where it is failing (am I missing a chat somewhere?), I can try to take a look.

Best

[somtooo]

@mstallone Yes this would work. but that issue is for people who use the github copilot provider not OpenAi as a provider.

[mstallone]

Ah, I understand now. I assumed copilot was just referencing the chat system, not GitHub Copilot.

I don't pay for that service, and I don't intend to at the moment, so if someone wants to take the lead on that, that'd be great. Probably Codex or Claude could easily adapt what I have without much effort.

[somtooo]

@mstallone all good pulled your branch and started adapting it for copilot already. Thanks for the initial work. But update commit that says closes issue #38858 😭 cause I was overjoyed in error when I read that before looking at the code changes

Edit: Was able to extend nicely and make the ResponseEventMapper.mapStream() shorter

[somtooo]

Looking at this code you pass in the stop sequences so you can handle it if you see it in the model response - (Correct me if im wrong). But when looking at the response api, vscode-chat implementation, openAi sdk i can find proof that the stop parameter is supported. (https://community.openai.com/t/how-to-prevent-the-api-returning-multiple-outputs/1251365/3).But even if it is i dont think the model response includes the actual stop sequences it returns the text before the stop sequence. Thats why the OpenAiEventMapper.map_stream function dosent check for it despite sending the stop sequence to the model.

If im correct then this code can be reduced which would be nice. Let me know thoughts. Thanks

[zed-industries-bot]

	Warnings
⚠️	This PR is missing release notes. Please add a "Release Notes" section that describes the change: Release Notes: - Added/Fixed/Improved ... If your change is not user-facing, you can use "N/A" for the entry: Release Notes: - N/A

Generated by 🚫 dangerJS against de0e5f2