DeepSeek R1 — For Software Development and API use
DeepSeek r1 vs OpenAI o1 Key Differences
The Internet and the stock market have been exploding (and imploding) since DeepSeek released its new thinking R1 model.
There are many myths and confusion around the hype — so I wanted to clarify and explain some key differences between OpenAI’s o1 model and the R1 model from DeepSeek.
Subscribe to our AI Growth Guys YouTube Channel
DeepSeek has made headlines mainly because of how quickly and inexpensively it created a competitive model to OpenAI’s o1 thinking model.
Here are some of the important details and how and when you might want to use DeepSeek in your next software project.
I’ll give some insight on the open-source version, where to test it, as well as DeepSeek’s API.
👉 Sign up for our free 5-day AI course to grow 🚀 and earn💲👈 with AI
Running DeepSeek R1 Locally on Your Computer for Free
One of the largest misleading parts about DeepSeek is the ability to run the Open-source R1 model locally on your computer.
Running DeepSeek-R1 locally requires substantial storage and computational resources due to its size and complexity. Here’s a breakdown of the storage requirements:
1. DeepSeek-R1 Model:
Parameters: 671 billion
Activated Parameters per Token: 37 billion
Storage Format: BF16 (16-bit floating point)
Estimated Storage Requirement:
Per Parameter Storage: 2 bytes (BF16)
Total Storage: 671 billion parameters × 2 bytes = approximately 1.34 terabytes
Realistically, if you want to use the R1 model for most people — you will need to use a Distilled version.
DeepSeek-R1-Distill-Qwen Series:
1.5B Parameters: Approximately 3 GB
7B Parameters: Approximately 14 GB
14B Parameters: Approximately 28 GB
32B Parameters: Approximately 64 GB
DeepSeek-R1-Distill-Llama Series:
8B Parameters: Approximately 16 GB
70B Parameters: Approximately 140 GB
The above sizes are much more realistic for a normal computer to handle. But these versions of the R1 model may not be nearly as good as the full version. They are certainly worth testing but its important to consider which version you are going to try on your computer.
Even if you have the storage space for the Full R1 model — your computer will most likely be way too underpowered without your own powerful GPU or graphics card.
Using DeepSeek R1 on a GPU Inference
A more viable option for most users is to try DeepSeek directly on Hugging Face or another Inference.
Running DeepSeek-R1 Without Local Resources
Hugging Face:
Host and access DeepSeek-R1 models online.
GPU Instances:
Use cloud-based GPUs for sufficient power and storage.
Why use an Infercence to Run DeepSeek-R1?
Avoid high local storage demands.
Access powerful hardware for smooth performance.
Test it fully to compare between OpenAI’s o1
If you don’t want to bother with either of these Options, DeepSeek also has an API for your to use an test.
This is the best option to test quickly but has its limitations.
Using the DeepSeek API
A great option to try DeepSeek for software development — is through their API.
If you are already using OpenAI’s API, you may want to switch to DeepSeek.
Here is an example of the potential cost savings of an application.
I asked ChatGPT-4o with web-search, to give a realistic example of potential cost-savings.
API Cost Comparison: DeepSeek R1 vs OpenAI o1 (Theoretical Example from ChatGPT-4o)
DeepSeek R1 API Pricing
Input Tokens:
Cache Hit: $0.14 per million
Cache Miss: $0.55 per million
Output Tokens: $2.19 per million
OpenAI o1 API Pricing
Input Tokens: $5.00 per million
Output Tokens: $15.00 per million
Cost Comparison Example
DeepSeek R1: $1.645 for 1M input (cache miss) + 500K output tokens
OpenAI o1: $12.50 for the same workload
Conclusion
DeepSeek R1 Total Cost: $1.65
OpenAI o1 Total Cost: $12.50
DeepSeek R1 is 7.5x more cost-effective than OpenAI o1 for similar workloads in this example — according to ChatGPT.
DeekSeek API is compatible with OpenAI’s
Below is a screenshot of configuration of DeepSeek’s API — if you already have an OpenAI API key.
You bascially only need to switch the API keys and url’s to make it work out of the box — a huge time saver.
But this is a little misleading because OpenAI has a far more robust API compared to DeepSeek’s so it currently may not work in your software setup.
DeepSeek’s API doesn’t have advanced features like fine-tuning and embeddings.
But, depending on your software — you may just be able to use the DeepSeek API and get considerable cost savings for similar results.
DeepSeek vs OpenAI — API Differences
DeepSeek API
Strengths: Cost-effective, excels in reasoning, math, and programming.
Ideal Use Cases: Budget-sensitive projects and chat-based interactions.
OpenAI API
Strengths: Versatile with advanced features like fine-tuning and embeddings.
Ideal Use Cases: Diverse AI needs, such as image generation and complex data analysis.
I also asked ChatGPT to provide insights when to use each API and reasons.
I personally would test the API of DeepSeek for any use cases. But I think its interesting to learn what an AI thinks and why.
Especially since DeepSeek is a direct competitor to ChatGPT.
DeepSeek API — Use Case Examples
1. Educational Tools
Example: A tutoring system that solves math problems and assists with coding exercises.
Why DeepSeek? Cost-efficient and excels at reasoning and problem-solving tasks.
2. Technical Support Chatbots
Example: A chatbot offering detailed troubleshooting for software and technical issues.
Why DeepSeek? Ideal for chat-based interactions and complex reasoning at a lower cost.
OpenAI API — Use Case Examples
1. Content Creation
Example: Writing blog posts, articles, or marketing materials.
Why OpenAI? Superior text generation and human-like creativity.
2. Image Generation
Example: Creating custom product visuals with DALL-E.
Why OpenAI? DeepSeek lacks image generation capabilities.
3. Language Translation
Example: Real-time multilingual translation for global applications.
Why OpenAI? Advanced language processing beyond DeepSeek’s scope.
Limitations:
Higher cost compared to DeepSeek.
If you enjoyed this article, have a look at my other channels below.
I’ll also be doing full DeepSeek API examples for software development soon — so stay tuned
I just started an AI Image generation platform Cartario — Train a custom AI model on your own photos or product photos to create the ultimate brand.
Try out Cartario for free (we just launched this and people are loving it!)
You can check out our AI podcast here.
My other Channels:
👉 Sign up to our free 5-Day email course to grow 🚀 and earn💲in the AI age
You can also sign up for my newsletter on how to use AI to earn more money.
Check out our YouTube Channel
Follow us at our website: AI Growth Guys