Skip to content

DeepSeek

DeepSeek offers high-quality models at extremely competitive prices, making it ideal for cost-sensitive production workloads.

Stable

Quick Facts

Property Value
Pricing Ultra-low cost ($0.07-0.27/1M tokens)
Models DeepSeek V3, DeepSeek Coder, DeepSeek R1
Streaming Supported
Functions Supported
Embeddings Supported
JSON Mode Supported

Setup

Get API Key

  1. Go to platform.deepseek.com
  2. Create an account
  3. Get API key from dashboard

Configure

export DEEPSEEK_API_KEY="..."

from sentimatrix.config import SentimatrixConfig, LLMConfig

config = SentimatrixConfig(
    llm=LLMConfig(
        provider="deepseek",
        api_key="...",
        model="deepseek-chat"
    )
)

llm:
  provider: deepseek
  model: deepseek-chat

Available Models

Model Input Cost Output Cost Best For
deepseek-chat $0.07/1M $0.27/1M General chat
deepseek-coder $0.07/1M $0.27/1M Code generation
deepseek-reasoner $0.55/1M $2.19/1M Complex reasoning (R1)

Usage Examples

Basic Usage

import asyncio
from sentimatrix import Sentimatrix
from sentimatrix.config import SentimatrixConfig, LLMConfig

config = SentimatrixConfig(
    llm=LLMConfig(
        provider="deepseek",
        model="deepseek-chat"
    )
)

async def main():
    async with Sentimatrix(config) as sm:
        # Extremely cost-effective summarization
        summary = await sm.summarize_reviews(reviews)
        print(summary)

asyncio.run(main())

Cost Comparison

# Analyze 10 million reviews
# DeepSeek: ~$5-10
# GPT-4o-mini: ~$10-15
# GPT-4o: ~$175+
# Claude 3.5: ~$250+

config = SentimatrixConfig(
    llm=LLMConfig(provider="deepseek", model="deepseek-chat")
)

Configuration Options

LLMConfig(
    provider="deepseek",
    model="deepseek-chat",

    # Generation settings
    temperature=0.7,
    max_tokens=4096,
    top_p=0.9,

    # Reliability
    timeout=60,
    max_retries=3,
)

Best Practices

  1. Use for High-Volume Workloads

    • Best cost-to-quality ratio
    • Great for production at scale

  2. Use deepseek-reasoner for Complex Tasks

    • R1 model for reasoning
    • Higher cost but better quality

  3. Combine with Fallbacks

    • Use DeepSeek as primary
    • Fall back to Groq/OpenAI