Run DeepSeek R1 671B via OpenRouter
Published on January 29, 2025   •   3 min read

Run DeepSeek R1 671B via OpenRouter

Jarek CeborskiJarek Ceborski

DeepSeek R1 is a groundbreaking new reasoning model that matches the performance of OpenAI's latest models while being fully open-source. What makes it truly special is its massive scale - it's a 671B parameter model, though it cleverly uses only 37B parameters during actual usage through an innovative architecture. In simple terms, it has the knowledge base of a massive model but runs with the efficiency of a smaller one.

OpenRouter↗ is a unified API gateway that lets you access multiple AI models through a single interface, including the full range of DeepSeek models. It currently supports multiple versions of DeepSeek R1: the Distill Llama 70B version, DeepSeek R1 (free tier), standard DeepSeek R1, DeepSeek R1 (nitro) for faster inference, and DeepSeek V3.

Kerlig Settings showing available DeepSeek R1 models via OpenRouter

How to Run DeepSeek R1 via OpenRouter in Kerlig

Here's how to get started with DeepSeek R1 in Kerlig:

  1. Get an OpenRouter API Key

    • First, visit the OpenRouter.ai↗ website
    • Click the Sign in button in the top right corner and then Sign up to create your account
    • Once signed in, go to API Keys↗ section.
    • Copy your API key

  2. Configure Kerlig

    • Open Kerlig on your Mac
    • Go to Settings ⌘, → Integrations
    • Select OpenRouter integration
    • Paste your OpenRouter API key
    • Click Save
    • Scroll DeepSeek Models section and toggle on the ones you want to use
💡Note
For the full 671B model, you need to select DeepSeek R1 or DeepSeek R1 (nitro)
  1. Run DeepSeek R1
    • Open Kerlig
    • Enter your prompt - any question you want to ask
    • Select one of newly added DeepSeek models
    • Click Run or press Enter

The integration with OpenRouter means you can easily switch between different versions of DeepSeek R1 depending on your needs - from the more affordable distilled version to the full 671B parameter model for the most demanding tasks.

For MacOS users, Kerlig makes it incredibly simple to harness the power of DeepSeek R1 for various tasks like email writing, Slack responses, Jira ticket creation, and document analysis. With support for over 350 models through integrations like OpenRouter, you can always choose the right tool for your specific needs.

Conclusion

Remember that while the full 671B DeepSeek R1 model is impressive in scale, it's designed to be efficient - using only 37B parameters during actual operation means you get the benefits of a massive model without the full computational overhead. This makes it practical for real-world usage while still maintaining state-of-the-art performance.

Download Kerlig↗ today and experience the power of one of the most advanced open-source language models available!