DeepSeek API Integration

Available Models

DeepSeek-V3 (deepseek-chat)

Generalist model with 671B parameters trained on 15 trillion tokens. Uses Mixture-of-Experts architecture with 128K context window.

  • • API Name: deepseek-chat
  • • 671B parameters, 37B activated per token
  • • $0.27 per million input tokens, $1.10 per million output tokens
  • • 128K context window

DeepSeek-R1 (deepseek-reasoner)

Advanced reasoning model designed for complex math, coding, and logical tasks. Competes directly with OpenAI's o1 model.

  • • API Name: deepseek-reasoner
  • • 671B parameters with reasoning capabilities
  • • Advanced reasoning for complex tasks
  • • 128K context window

DeepSeek-R1-0528

Updated R1 model with system prompts, JSON output, and function calling support for agentic AI use cases.

  • • API Name: deepseek-r1-0528
  • • System prompt support
  • • JSON output and function calling
  • • Released May 2025

DeepSeek-V3-0324

Improved version of V3 with enhanced performance. Updated weights for better quality and capabilities.

  • • API Name: deepseek-v3-0324
  • • Improved V3 weights
  • • Enhanced performance metrics
  • • Updated March 2025

⚠️ Important Notice

Information about models, pricing, and features may be outdated or incorrect. Always consult the official provider documentation for the most current and accurate data.

Key Features

Cutting-Edge Architecture

DeepSeek-V3 uses Mixture-of-Experts (MoE) with Multi-head Latent Attention (MLA) for efficient training and inference.

  • • 671B total parameters, 37B activated
  • • MLA and DeepSeekMoE architectures
  • • Only 2.788M H800 GPU hours for training
  • • Exceptional performance metrics

Performance Benchmarks

Leading performance across various benchmarks including MMLU (87.1%), BBH (87.5%), and mathematical reasoning tasks.

  • • MMLU: 87.1% accuracy
  • • BBH: 87.5% performance
  • • Outperforms open-source models
  • • Comparable to leading closed-source models

API Compatibility

OpenAI-compatible API format makes integration straightforward for developers familiar with OpenAI tools.

  • • OpenAI SDK compatibility
  • • Easy migration from other providers
  • • Standard REST API endpoints
  • • Comprehensive documentation

Open Source & Cost-Effective

Available under open-source license with significantly lower pricing compared to competitors like GPT-4o.

  • • Open-source model weights
  • • Cheaper than average pricing
  • • $0.48 per 1M tokens (blended 3:1)
  • • No vendor lock-in

Technical Specifications

  • • Secure API key storage with iOS Keychain
  • • Native iOS integration
  • • Support for streaming responses
  • • Automatic token counting
  • • Rate limiting and monitoring
  • • Error handling and retry logic