Cursor Model Selection & Configuration

Available Models

Built-in Models (Cursor Subscription)

Using Your Own API Keys

// Settings > Models > API Keys
{
  "openai": {
    "apiKey": "sk-...",
    "organization": "org-..." // optional
  },
  "anthropic": {
    "apiKey": "sk-ant-..."
  },
  "azure": {
    "apiKey": "...",
    "endpoint": "https://your-resource.openai.azure.com",
    "deployment": "gpt-4"
  },
  "google": {
    "apiKey": "...",
    "model": "gemini-pro"
  }
}

Model Selection by Task

Completions (Tab)

Recommended: GPT-3.5 Turbo or Cursor-small
- Fast response time critical
- Context window usually sufficient
- Cost effective for high volume

Settings:
{
  "cursor.completion.model": "gpt-3.5-turbo"
}

Chat

Task-based selection:

Quick questions → GPT-3.5 Turbo
Code review → GPT-4 Turbo
Explanations → Claude 3.5 Sonnet
Complex architecture → GPT-4 or Claude Opus
Debugging → GPT-4 Turbo (large context)

Composer

Recommended: GPT-4 Turbo or Claude 3.5 Sonnet
- Multi-file edits need accuracy
- Large context for project understanding
- Quality over speed

Settings:
{
  "cursor.composer.model": "gpt-4-turbo"
}

Switching Models

In Chat

Method 1: Model dropdown in chat header
Method 2: Type model name in chat:
  "Using GPT-4: explain this code"
Method 3: Settings > Default Chat Model

Per-Conversation

Start chat with model preference:
"[Claude] Explain this authentication flow"
"[GPT-4] Review this for security issues"

Global Defaults

// settings.json
{
  "cursor.chat.defaultModel": "gpt-4-turbo",
  "cursor.completion.model": "gpt-3.5-turbo",
  "cursor.composer.model": "gpt-4-turbo"
}

Model Comparison

GPT-4 vs Claude 3.5 Sonnet

GPT-4 Turbo:
✓ Better at code generation
✓ Stronger at following complex instructions
✓ More consistent output format
✗ Can be verbose

Claude 3.5 Sonnet:
✓ Better explanations
✓ More natural conversation
✓ Excellent at documentation
✗ Sometimes over-cautious

Speed vs Quality Tradeoffs

Speed Priority (GPT-3.5 Turbo):
- Autocomplete while typing
- Quick lookups
- Simple refactors
- ~100-500ms response

Quality Priority (GPT-4):
- Architecture decisions
- Security reviews
- Complex debugging
- ~2-10s response

Cost Optimization

Cursor Subscription

Free: 2000 completions, 50 slow requests/month
Pro ($20/mo): Unlimited completions, 500 fast requests
Business ($40/user): Everything + admin features

Fast vs Slow:
- Fast: GPT-4 Turbo, Claude Opus
- Slow: Queue-based, may take 30s+

Using Own API Keys

Cost comparison (approx per 1M tokens):
GPT-4 Turbo: $10 input, $30 output
GPT-3.5 Turbo: $0.50 input, $1.50 output
Claude Sonnet: $3 input, $15 output
Claude Opus: $15 input, $75 output

Strategy:
- Use GPT-3.5 for completions (high volume)
- Use GPT-4 for chat (lower volume, higher stakes)

Model-Specific Settings

OpenAI Models

{
  "cursor.openai.temperature": 0.7,
  "cursor.openai.maxTokens": 4096,
  "cursor.openai.topP": 1.0,
  "cursor.openai.frequencyPenalty": 0,
  "cursor.openai.presencePenalty": 0
}

Anthropic Models

{
  "cursor.anthropic.temperature": 0.7,
  "cursor.anthropic.maxTokens": 4096
}

Advanced Configuration

Custom Endpoints

// For self-hosted or proxy setups
{
  "cursor.api.baseUrl": "https://your-proxy.com/v1",
  "cursor.api.headers": {
    "X-Custom-Header": "value"
  }
}

Model Fallbacks

// If primary fails, use fallback
{
  "cursor.model.fallback": {
    "gpt-4": "gpt-3.5-turbo",
    "claude-3-opus": "claude-3-sonnet"
  }
}

Troubleshooting Models

"Model Not Available"

1. Check subscription tier
2. Verify API key if using own
3. Check model name spelling
4. Some models require waitlist

"Rate Limited"

1. Switch to different model
2. Wait for limit reset
3. Use own API key
4. Upgrade plan

"Context Too Long"

1. Use larger context model (GPT-4 Turbo)
2. Reduce selected code
3. Start new conversation
4. Be more concise in prompts