CCProxy

Appearance

DeepSeek Provider ​

Overview ​

DeepSeek is an advanced AI provider offering state-of-the-art language models with exceptional reasoning capabilities, competitive coding performance, and significant improvements in accuracy and reliability.

⚠️ Important Limitations ​

  • NO Function Calling Support: The DeepSeek transformer in CCProxy does not support function calling/tools. This makes DeepSeek incompatible with Claude Code, which requires function calling for its operations. Only use DeepSeek through CCProxy for direct API calls that don't require tools.
  • Max Tokens Hard Limit: DeepSeek enforces a hard limit of 8192 max_tokens. Any request exceeding this will be automatically capped at 8192.
  • Claude Code Compatibility: Due to lack of function calling support, DeepSeek is NOT recommended for use with Claude Code. Consider using Anthropic, OpenAI, or Gemini providers instead.

Latest Models (July 2025) ​

Available Models ​

  • deepseek-chat (maps to DeepSeek-V3-0324) - Latest V3 with improved post-training
  • deepseek-reasoner (maps to DeepSeek-R1-0528) - Advanced reasoning model with built-in reasoning capabilities
  • deepseek-coder-v2 - Specialized for coding tasks

Key Features ​

DeepSeek-R1-0528 (deepseek-reasoner) ​

  • 87.5% on AIME 2025 (up from 70%) - Exceptional mathematical reasoning
  • 45-50% less hallucination - Significantly improved reliability
  • Advanced multi-step reasoning capabilities via reasoning_content field
  • Optimal for complex problem-solving tasks
  • Note: While this model has excellent reasoning capabilities, its limited function calling support may impact Claude Code compatibility

DeepSeek-V3-0324 (deepseek-chat) ​

  • Better reasoning and analytical capabilities
  • Enhanced coding skills with improved debugging
  • Basic tool-use capabilities (limited compared to other providers)
  • General-purpose model with balanced performance

DeepSeek-Coder-V2 ​

  • Specialized for software development tasks
  • Superior code generation and understanding
  • Multi-language programming support
  • Optimized for technical documentation

Technical Specifications ​

  • Supports system prompts natively
  • Built-in reasoning capabilities via reasoning_content field (no need for <think> tags)
  • Streaming support for real-time responses with special handling for reasoning content
  • JSON mode support for structured outputs
  • Max tokens: Hard-capped at 8192 tokens

Configuration ​

Basic Configuration ​

json
{
  "providers": {
    "deepseek": {
      "apiKey": "${DEEPSEEK_API_KEY}",
      "baseURL": "https://api.deepseek.com/v1",
      "models": {
        "default": "deepseek-chat",
        "think": "deepseek-reasoner",
        "code": "deepseek-coder-v2"
      }
    }
  }
}

Environment Variables ​

bash
export DEEPSEEK_API_KEY="your-api-key-here"
export CCPROXY_CONFIG="/path/to/config.json"

Routing Recommendations ​

Default Route ​

Use deepseek-chat for general-purpose tasks:

  • Conversational AI
  • Content generation
  • Basic reasoning tasks
  • Tool use and function calling

Think Route ​

Use deepseek-reasoner for complex reasoning:

  • Mathematical problems
  • Logic puzzles
  • Multi-step reasoning
  • Scientific analysis
  • Strategic planning

Code Route ​

Use deepseek-coder-v2 for development tasks:

  • Code generation
  • Code review and debugging
  • Technical documentation
  • API design
  • Algorithm implementation

Integration Examples ​

Claude Code Configuration ​

json
{
  "routes": {
    "default": {
      "provider": "deepseek",
      "model": "deepseek-chat"
    },
    "think": {
      "provider": "deepseek",
      "model": "deepseek-reasoner"
    },
    "longContext": {
      "provider": "deepseek",
      "model": "deepseek-chat"
    }
  }
}

API Request Example ​

bash
curl -X POST http://localhost:3456/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: your-ccproxy-key" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-3-5-sonnet-20241022",
    "messages": [{"role": "user", "content": "Explain quantum computing"}],
    "max_tokens": 1024
  }'

CCProxy will automatically route to DeepSeek based on your configuration.

Streaming Example ​

javascript
const response = await fetch('http://localhost:3456/v1/messages', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'x-api-key': 'your-ccproxy-key',
    'anthropic-version': '2023-06-01'
  },
  body: JSON.stringify({
    model: 'claude-3-5-sonnet-20241022',
    messages: [{ role: 'user', content: 'Write a Python function' }],
    stream: true
  })
});

const reader = response.body.getReader();
// Handle streaming response

Advanced Configuration ​

Model-Specific Parameters ​

json
{
  "providers": {
    "deepseek": {
      "apiKey": "${DEEPSEEK_API_KEY}",
      "models": {
        "default": "deepseek-chat",
        "think": "deepseek-reasoner",
        "code": "deepseek-coder-v2"
      },
      "parameters": {
        "temperature": 0.7,
        "top_p": 0.95,
        "frequency_penalty": 0,
        "presence_penalty": 0
      }
    }
  }
}

Multi-Route Configuration ​

json
{
  "routes": {
    "math": {
      "provider": "deepseek",
      "model": "deepseek-reasoner",
      "parameters": {
        "temperature": 0.1
      }
    },
    "creative": {
      "provider": "deepseek", 
      "model": "deepseek-chat",
      "parameters": {
        "temperature": 0.9
      }
    },
    "debug": {
      "provider": "deepseek",
      "model": "deepseek-coder-v2",
      "parameters": {
        "temperature": 0
      }
    }
  }
}

Performance Considerations ​

Model Selection ​

  • deepseek-reasoner: Best for accuracy-critical tasks, slightly higher latency
  • deepseek-chat: Balanced performance for most use cases
  • deepseek-coder-v2: Optimized for code generation speed

Optimization Tips ​

  1. Use appropriate models for specific tasks
  2. Adjust temperature based on task requirements
  3. Enable streaming for better perceived performance
  4. Consider token limits for cost optimization

Troubleshooting ​

Common Issues ​

API Key Errors

bash
# Verify API key is set
echo $DEEPSEEK_API_KEY

# Test direct API access
curl https://api.deepseek.com/v1/models \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY"

Model Not Found

  • Ensure you're using the correct model names: deepseek-chat, deepseek-reasoner, or deepseek-coder-v2
  • Check CCProxy configuration for proper model mapping

Rate Limiting

  • DeepSeek has rate limits per API key
  • Implement retry logic with exponential backoff
  • Consider using multiple API keys for high-volume applications

Best Practices ​

  1. Model Selection: Choose models based on task requirements
    • For Claude Code: Consider using other providers with better function calling support
    • For reasoning tasks without tool use: deepseek-reasoner is excellent
    • For general chat: deepseek-chat works well within its limitations
  2. Temperature Settings: Lower for factual tasks, higher for creative tasks
  3. System Prompts: Leverage native system prompt support for consistent behavior
  4. Token Limits: Be aware of the 8192 token hard limit when designing prompts
  5. Error Handling: Implement robust error handling for API failures and function calling limitations
  6. Monitoring: Track usage and performance metrics, especially tool use failures

When to Use DeepSeek ​

Good Use Cases:

  • Mathematical reasoning and problem solving
  • Code generation without extensive tool use
  • General conversation and Q&A
  • Tasks requiring deep reasoning but not function calling

Consider Alternatives When:

  • Using Claude Code with its extensive function calling requirements
  • Needing reliable tool/function calling capabilities
  • Requiring more than 8192 max tokens
  • Building applications heavily dependent on tool use

See Also ​