Allow disabling Responses API for OpenAI-compatible servers with gpt-5-like model aliases

Before submitting your bug report

• I've tried finding an answer on the Continue docs site
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

- OS: Linux CachyOS
- Continue version: 2.0
- IDE version: Rider 2026.1.4
- Model: openai
- config:
  
name: Local Config
version: 1.0.0
schema: v1

models:
  - name: gpt-5.5
    provider: openai
    model: gpt-5.5
    apiBase: https://chatgpt.myCompany.de/api/
    apiKey: xxxx
    useLegacyCompletionsEndpoint: false
 

Description

Problem

When using the openai provider with a custom OpenAI-compatible API and a model named gpt-5.5, Continue automatically uses the Responses API endpoint:

POST /responses

This fails because the backend only supports OpenAI-compatiblnAI provider forces /responses for custom gpt-5-like model names on OpenAI-compatible APIse Chat Completions:

POST /chat/completions

The model name is a custom/internal alias and does not imply that the backend supports the OpenAI Responses API.

Evidence

This fails:

curl -i https://chatgpt.myCompany.de/api/responses
-H "Authorization: Bearer xxx"
-H "Content-Type: application/json"
-d '{"model":"gpt-5.5","input":"Hallo"}'

Result:

HTTP 405 Method Not Allowed

This works:

curl -i https://chatgpt.myCompany.de/api/chat/completions
-H "Authorization: Bearer xxx"
-H "Content-Type: application/json"
-d '{
"model": "gpt-5.5",
"messages": [{"role": "user", "content": "Hallo"}]
}'

Result:

HTTP 200 OK

https://github.com/continuedev/continue/blob/d0a3c0b626b5bebc3bef4742eec05a0242be0bab/core/llm/llms/OpenAI.ts#L221-L223

Expected behavior

There should be a config option to disable the Responses API for OpenAI-compatible backends, e.g.

useResponsesApi: false

or Continue should only use /responses automatically for the official OpenAI API, not for arbitrary apiBase values.

useLegacyCompletionsEndpoint: false does not solve this, because the issue is not /completions vs /chat/completions, but /responses vs /chat/completions.

To reproduce

No response

Log output

Start in core/llm/llms/OpenAI.ts at the linked lines where gpt-5-like model names select the Responses API. Trace how the OpenAI model configuration is read, then determine where an option such as useResponsesApi could be supported for custom apiBase values. Done means a custom OpenAI-compatible backend can use /chat/completions without being forced to /responses.