Custom DeepSeek API integration causes "Provider Error" on native Cursor models

Hi Cursor Team,

I am experiencing an issue where configuring a custom API key for DeepSeek seems to break the native Cursor models.

Environment:

  • OS: Windows

  • Mode: Ask and Agent Mode

  • Cursor Version: Version 3.19.7

Steps to Reproduce:

  1. I added my DeepSeek API key in Settings > Models > API Keys.

  2. I enabled “Override OpenAI Base URL” and set it to https://api.deepseek.com/v1.

  3. I verified my DeepSeek key is valid by successfully calling https://api.deepseek.com/models via curl.

  4. I manually added the model IDs deepseek-flash and deepseek-v4-pro.

  5. I selected a native Cursor model, specifically Grok 4.7 High Fast.

The Issue:
When I try to send a simple message (e.g., “hello”) using the native Grok 4.7 High Fast model, I receive the following error:

Provider Error
We’re having trouble finding the resource you requested. If the problem persists, please contact support.
Request ID: bad63ae2-f7e6-4cbb-b1cf-eb6e66e2e202

(You can also see the “Invalid API key” error when I first tried to use the DeepSeek models, but it seems the request is now just routing incorrectly.)

Questions:
It seems that adding a custom DeepSeek endpoint/key is somehow interfering with the routing of native Cursor models. Is this a known issue with the new API key configuration UI? How can I use my custom DeepSeek key without breaking the native Cursor models (like Grok)?

Thank you for your help!

Hey @ye_chen!

Quick fix: in Cursor Settings > Models > API Keys, turn off the OpenAI API Key toggle when you want native models (your key and base URL stay saved), start a new chat, and Grok 4.7 High Fast should run on your Cursor plan again. Flip it back on when you want DeepSeek.

Why it happens: the OpenAI key + Override OpenAI Base URL currently applies to every model that isn’t Claude or Gemini, so your Grok request was getting routed to DeepSeek. There’s no per-model scoping yet, and it’s an issue we’re tracking.

One more thing: your DeepSeek “model not found” usually means the IDs don’t match exactly. Check they’re spelled exactly as the id values from https://api.deepseek.com/models, and if they still fail, reply with the exact name and a fresh Request ID.