DeepSeek R1

A Powerful Reasoning Model You Can Use Today

PublishedJanuary 27, 2025

•2 min read•View as Markdown

Bakunga Bronson

Comments

Join the discussion

No comments yet. Be the first to comment.

More from this blog

The Battle of Galen Erso Recap

Insights gained from the battle.

Feb 4, 20255 min read

GitHub Models: Easy Setup Instructions

Guide to Using GitHub Models with OpenAI-Compatible Software

Dec 8, 20242 min read

Claude Desktop Model Context Protocol Setup

Enhance Claude.ai's abilities with MCP

Dec 4, 20242 min read

Claude Desktop Model Context Protocol Setup

Anthropic Model Context Protocol (MCP).

Understanding What Lies Ahead.

Nov 28, 20245 min read

Bakunga Bronson's Blog

7 posts

How to Access DeepSeek R1

Web App: The easiest way to try DeepSeek is through their web application at chat.deepseek.com. You can use the V3 model for general tasks and switch to the R1 model for tasks requiring more thinking. Note: While the chat interface is free, API access might involve fees.

Ollama: For more advanced users, DeepSeek R1 can be run locally using Ollama. Visit ollama.com, download the Ollama software for your operating system (macOS, Linux, or Windows), and then select the desired DeepSeek R1 model size from the available options (1.5B, 7B, 14B, 32B, 70B, or 67B parameters). Each model has different RAM requirements, so choose one that's appropriate for your system.

Using DeepSeek R1 with Other Models

A recommended workflow is to use DeepSeek R1 first to plan or strategise the solution to a complex problem. Then, feed the output from DeepSeek R1 to a general-purpose large language model (LLM) like Claude or GPT to execute the plan. This approach can be more cost-effective than relying solely on more expensive LLMs for the entire process.

Cost Comparison

OpenAI models (e.g., GPT) can cost $10-16 per million output tokens.

Claude models can cost $15 per million output tokens.

DeepSeek R1, accessed via their API or OpenRouter, is significantly more affordable, costing around $2 per million output tokens. The free chat interface makes it even more attractive for experimentation.

I'd like to point out that pricing and features can change, so please always refer to the official sources for the most up-to-date information.

Links

DeepSeek Web App: chat.deepseek.com

DeepSeek Website (for API info, etc.): deepseek.com (Note: The API Platform link shown in the video resulted in a 404 error, suggesting a potential temporary issue.)

Hugging Face DeepSeek R1 Repository: huggingface.co/deepseek/DeepSeek-R1 (For downloading the model directly, though this is intended for more technical users).

OpenRouter (for accessing DeepSeek R1 and other LLMs): openrouter.ai

Ollama (for running LLMs locally): ollama.com

Arena (LLM Leaderboard): arena.lmsys.org (This was used for comparing model performance)

Command Palette