Quick answer: In Cursor open Settings → Models, paste your 321AI key under OpenAI API Key, enable Override OpenAI Base URL and set it to https://321ai.cc/v1 (the /v1 suffix is required), click Verify, then add model IDs such as gpt-5.5 or claude-sonnet-5. Tab completion uses Cursor's own service and is not billed through this key.
Audience: Cursor users who want their own billing endpoint for Chat and Composer with GPT / Claude / DeepSeek models.
| Setting | Value | Note |
|---|---|---|
OpenAI API Key | your 321AI key | Settings → Models |
Override OpenAI Base URL | https://321ai.cc/v1 | must include /v1 |
Model IDs | gpt-5.5 / claude-sonnet-5 / … | manual list, exact names |
Double-check that the base URL includes /v1, the key is valid and has quota, and the model name exists on 321AI.
No. Tab autocomplete is part of Cursor's built-in service; only requests from Chat/Composer and similar features go through your custom endpoint — check Cursor's official docs for the current scope.
Configure a cheap model (e.g. a deepseek or glm variant) for long background chats, and check per-model prices on /pricing/openai before heavy usage.
Related: 中文版 Cursor 教程 · Claude Code setup · price table