Cursor
Use MiniMax M3.1 Flash in Cursor
Cursor talks to the OpenAI-compatible path; the rest is field-by-field configuration and a verification step.
How this works
Cursor lets you register a custom model against an OpenAI-compatible endpoint. MiniMax exposes that shape at https://api.minimax.io/v1, so the model can be added as a custom provider rather than through a marketplace integration.
Setup steps
- Open Settings → Models in Cursor.
- Choose OpenAI API Key as the provider type (the option that accepts a custom base URL).
- Paste the MiniMax API key.
- Set the base URL to
https://api.minimax.io/v1— note this is the OpenAI-compatible path, not the Anthropic-compatible/anthropicone used by Claude Code. - Add the model id
MiniMax-M3.1-Flash. - Save, then select the model in a new chat.
Which endpoint to use
| Client | Endpoint |
|---|---|
| Claude Code | https://api.minimax.io/anthropic |
| Codex, Cursor, generic OpenAI clients | https://api.minimax.io/v1 |
| Ollama cloud | https://ollama.com with minimax-m3.1-flash:cloud |
Mixing these up is the most common cause of a 404 or an “invalid model” response that looks like a bad key.
Verify the connection
- Send a one-line message and check the model name Cursor reports for the request.
- Ask the model something with a known answer and confirm the response is coherent; a misrouted request to the wrong provider usually shows up as an empty or oddly formatted reply.
- For a stronger check, attach a small image: M3.1 Flash accepts image input, and a correct model will describe it.
- Watch the token or latency behaviour. Long responses are expected, because reasoning is always on.
Troubleshooting
| Symptom | Cause | Fix |
|---|---|---|
| Authentication error | Wrong key type, or the key belongs to a subscription rather than the platform | Use a platform API key from the developer console |
| Model not listed after saving | Entry not committed, or an invalid model id | Re-add with MiniMax-M3.1-Flash |
| 404 or unknown route | Base URL pointing at /anthropic | Use /v1 |
| Very slow first token | Reasoning always on, effort at max | Lower the effort in the API layer if you control it |
| Long context rejected | Context beyond the harness limit | Check the model context setting in Cursor |
When to use MiniMax Code instead
Cursor is an editor-first agent; MiniMax Code is MiniMax’s own desktop agent with the model pre-wired and no key setup if you are on an M Plan subscription. If your work is long multi-step automation inside one app rather than editing a codebase, MiniMax Code is the more direct path. If you live in an IDE, Cursor is the better fit. The M Plan page covers the subscription side.
What we could not verify
We have not published a step-by-step Cursor reproduction of this configuration with our own screenshots. The steps above follow MiniMax’s own Cursor documentation and its general OpenAI-compatible custom-model flow; treat the interface paths as subject to change and tell us if they no longer match — the contact page has a correction route.
Next: OpenCode and other terminal agents, or the API reference if you are wiring it up yourself.
Sources and version notes
Checked October 21, 2026. We label official claims, third-party measurements, our own testing and community reports separately — see the sources policy for what each label means.