DeepSeek

DeepSeek develops highly capable open-source models that compete with frontier closed models, particularly excelling at coding, math, and reasoning tasks at dramatically lower cost. DeepSeek made waves with models that match GPT-4-class performance while being fully open-weight.

Key use cases include:

• Cost-effective API alternative to GPT-4 and Claude for production workloads
• Coding assistance with DeepSeek-Coder, competitive with Copilot-class models
• Mathematical and scientific reasoning with DeepSeek-R1
• Self-hosted deployments with full control over data and cost
• Drop-in replacement in OpenAI-compatible applications
• Research and fine-tuning with fully open weights

DeepSeek is popular among cost-conscious teams and researchers who want frontier-level capabilities without frontier-level pricing. The DeepSeek-V3 model offers an exceptional quality-to-cost ratio, and the API follows the OpenAI format for easy migration.

Models are available both through the hosted API and as open-weight downloads for self-hosting on your own infrastructure.

1. Sign up at platform.deepseek.com and generate an API key.
2. Use the OpenAI-compatible SDK:

   pip install openai

3. Point to the DeepSeek API:

   from openai import OpenAI
   client = OpenAI(
       api_key='sk-...',
       base_url='https://api.deepseek.com'
   )
   response = client.chat.completions.create(
       model='deepseek-chat',
       messages=[{'role': 'user', 'content': 'Hello!'}]
   )

4. For self-hosting, download from Hugging Face and serve with vLLM or Ollama.

Pricing: DeepSeek-V3 API is ~$0.27/1M input tokens and ~$1.10/1M output tokens — roughly 10-30x cheaper than GPT-4-class models. Open-weight models are free to self-host. See platform.deepseek.com.