[BUG] azure gpt-5.3-codex not working

[Raffa50]

Problem (one or two sentences)

Hello I'm trying to configure gpt-5.3-codex hosted on my azure subscription

Provider: openai compatible
Url: https://.openai.azure.com/openai/deployments/gpt5.3
model: gpt-5.3-codex
use Azure: true
enable R1 parameters: false (also tryed true)

I get the error:

OpenAI completion error: 400 The chatCompletion operation does not work with the specified model, gpt-5.3-codex. Please choose different model and try again. You can learn more about which models can be used with each operation here: https://go.microsoft.com/fwlink/?linkid=2197993.

Context (who is affected and when)

When you use an azure gpt-5.x codex model

Reproduction steps

1. Configure an azure gpt-5.x model
2. Ask anything in the chat

Expected result

Roo code works

Actual result

Error 400

Variations tried (optional)

No response

App Version

3.52.1

API Provider (optional)

OpenAI Compatible

Model Used (optional)

gpt-5.3-codex

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

I've analyzed this issue and here's my proposed implementation plan:

Root Cause: The OpenAiHandler (used by the "OpenAI Compatible" provider) exclusively uses client.chat.completions.create() for all API calls. Azure's GPT-5.x codex models (like gpt-5.3-codex) do not support the Chat Completions API -- they only support the Responses API. This is why the error says "The chatCompletion operation does not work with the specified model, gpt-5.3-codex."

The OpenAiNativeHandler already uses responses.create() for all models and handles codex models correctly. The "OpenAI Compatible" handler needs the same capability when a codex model is detected.

Proposed Fix:

1. Add codex model detection in src/api/providers/openai.ts - Add a helper method _isCodexModel(modelId) that checks if the model ID contains "codex" (e.g., gpt-5.3-codex, gpt-5.1-codex).

2. Add Responses API path in OpenAiHandler.createMessage() - When a codex model is detected (similar to how o1/o3/o4 models are already routed to handleO3FamilyMessage), route to a new handleCodexMessage() method that uses client.responses.create() instead of client.chat.completions.create().

3. Implement handleCodexMessage() in OpenAiHandler - This method will:

   • Format the conversation for the Responses API (using instructions for system prompt, input array for messages)
   • Use (this.client as any).responses.create(requestBody) to call the Responses API
   • Process the streaming response events (text deltas, tool calls, usage) similar to how OpenAiNativeHandler does it
   • Handle both streaming and non-streaming modes

4. Update completePrompt() as well - Add the same codex detection to the completePrompt method so simple completions also work.

5. Add tests - Add test cases in src/api/providers/__tests__/openai.spec.ts to verify that codex models are routed through the Responses API path.

This approach keeps the change minimal and scoped -- it only affects codex model requests while preserving existing behavior for all other models. It mirrors the pattern already used for o1/o3/o4 model detection in the same handler.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[Raffa50]

@roomote I approve

I've seen your approval and will proceed with the implementation. I'll open a PR to share progress shortly.

Opened PR #11952 with my attempt at addressing this. The fix detects codex models and routes them through the Responses API instead of the Chat Completions API. Feedback welcome!

[louyshong]

Could this PR be merged soon? We are really looking forward to able to use the latest codex models via Azure OpenAI...

[Raffa50]

It still not working with latest roo code 3.51.1:

Date/time: 2026-04-03T12:22:57.992Z
Extension version: 3.51.1
Provider: openai (proxy)
Model: gpt-5.3-codex

400
OpenAI completion error: 400 The requested operation is unsupported.
You seem to be using a proxy-based provider. Make sure to check its logs and to ensure it's not rewriting Roo's requests.

[Raffa50]

The issue persists with latest version: 3.52.1.
Any updates? does gpt-5.3-codex works for anybody?

[louyshong]

Codex models still don't work as they don't support the completions API. The latest GPT 5.4 on Azure OpenAI works with Roo Code though.

[Raffa50]

But codex models could be integrated too. As you see @roomote proposed a fix PR #11952

I've seen your comment @Raffa50. PR #11952 already has the fix for this -- all CI checks pass and it's mergeable. I'll rebase it on the latest main and mark it as ready for review to help move it along.