Cursor Custom AI Models: Connect Claude, Gemini, DeepSeek

Problem: Cursor Only Shows a Fixed Model List — You Want Your Own

Cursor ships with a curated model list (GPT-4o, Claude Sonnet, etc.), but you may need a model that isn't there — DeepSeek R1 for cost-efficient reasoning, Gemini 2.0 Flash for long context, or a specific Claude version you're already paying for via the Anthropic API.

The fix is Cursor's Custom Models feature. It lets you bring any OpenAI-compatible endpoint into the IDE.

You'll learn:

• How to add Claude, Gemini, and DeepSeek as custom models in Cursor
• The exact model strings and base URLs for each provider
• How to verify the models work and fix the most common errors

Time: 10 min | Difficulty: Beginner

Why Custom Models Exist in Cursor

Cursor routes all AI requests through its own proxy by default. When you add a custom model, you bypass the proxy for that model and hit the provider's API directly using your own key. That means:

• You control cost — charges go to your own API account, not Cursor's usage pool
• You get access to unreleased or niche models — anything the provider exposes via API
• Rate limits are yours — useful if you're on a high-tier API plan

The trade-off: custom models don't count toward Cursor's fast request quota, and some Cursor features (like auto-complete) only work with a subset of supported models.

Solution

Step 1: Open Cursor Model Settings

1. Open Cursor
2. Press Cmd + , (macOS) or Ctrl + , (Windows/Linux) to open Settings
3. Navigate to Models in the left sidebar

You'll see the default model list at the top and a + Add Model button below it.

Step 2: Add Claude (Anthropic)

Anthropic's API is not OpenAI-compatible natively, but Cursor handles it directly — no base URL override needed.

Model strings for Claude (copy exactly):

In Cursor Settings → Models:

1. Click + Add Model
2. Enter the model string (e.g. claude-sonnet-4-5)
3. Leave Base URL blank — Cursor uses the Anthropic endpoint automatically
4. Go to API Keys section in the same Settings panel
5. Paste your Anthropic API key into the Anthropic API Key field

API Key field: sk-ant-api03-...
Model string:  claude-sonnet-4-5
Base URL:      (leave empty)

Click Verify to confirm the key works. You should see a green checkmark.

Step 3: Add Gemini (Google)

Google exposes Gemini via an OpenAI-compatible endpoint. Use this base URL:

https://generativelanguage.googleapis.com/v1beta/openai/

Model strings for Gemini:

In Cursor Settings → Models:

1. Click + Add Model
2. Enter the model string (e.g. gemini-2.0-flash)
3. Set Base URL to https://generativelanguage.googleapis.com/v1beta/openai/
4. In the OpenAI API Key field, paste your Google AI Studio API key

API Key field: AIzaSy...
Model string:  gemini-2.0-flash
Base URL:      https://generativelanguage.googleapis.com/v1beta/openai/

Get your Google AI Studio key at aistudio.google.com/apikey.

Step 4: Add DeepSeek

DeepSeek's API is OpenAI-compatible and among the cheapest options for reasoning-class models.

Base URL: https://api.deepseek.com/v1

Model strings for DeepSeek:

In Cursor Settings → Models:

1. Click + Add Model
2. Enter deepseek-chat or deepseek-reasoner
3. Set Base URL to https://api.deepseek.com/v1
4. Paste your DeepSeek API key into the OpenAI API Key field

API Key field: sk-...
Model string:  deepseek-chat
Base URL:      https://api.deepseek.com/v1

Get your DeepSeek key at platform.deepseek.com.

Step 5: Select the Model in Chat or Composer

Once added, the model appears in the model picker dropdown in:

• Chat panel — click the model name at the bottom of the chat input
• Composer — same dropdown in the Composer toolbar
• Inline edit (Cmd+K) — model selector in the floating bar

Custom models are marked with a small person icon (indicating they use your API key).

Verification

Open the Chat panel, select your new model, and send a test prompt:

What model are you and who made you?

You should see: The model correctly identifying itself — e.g. "I'm Claude, made by Anthropic" or "I'm Gemini, a large language model from Google."

If the model selector shows the name but responses fail, check the troubleshooting section below.

Troubleshooting

Error: Invalid API Key

• Double-check the key has no trailing spaces
• Confirm the key is active in the provider's dashboard
• For Gemini: make sure you're using an AI Studio key, not a Google Cloud Vertex key

Error: Model not found

• The model string is case-sensitive — copy it exactly from the tables above
• For Gemini: confirm the model is available in your region at ai.google.dev/gemini-api/docs/models

Responses are very slow

• DeepSeek Reasoner and Claude Opus are inherently slower — switch to Flash or Sonnet for everyday tasks
• Check provider status pages for ongoing incidents

Model appears in list but doesn't show in Composer

• Restart Cursor after adding models — the Composer model list sometimes needs a reload
• Confirm draft: false — no, wrong checklist. Confirm the model is toggled ON in the Models list (the toggle to the left of the model name)

What You Learned

• Cursor supports any OpenAI-compatible API endpoint via the custom model system
• Claude uses Cursor's native Anthropic integration — no base URL needed
• Gemini and DeepSeek both use OpenAI-compatible endpoints with different base URLs
• Custom models bill directly to your API account, not Cursor's usage quota

When to use each model in Cursor:

• Claude Sonnet — best overall for code generation and refactoring
• Gemini 2.0 Flash — fastest option, good for large file context (1M token window)
• DeepSeek Chat — most cost-effective for high-volume usage
• DeepSeek Reasoner — step-by-step debugging of tricky logic bugs

Tested on Cursor 0.48, macOS Sequoia 15.3 and Windows 11