In today’s rapidly developing artificial intelligence landscape, a major breakthrough has quietly emerged – Chinese AI company DeepSeek has released their fully open-source DeepSeek R1 model, which rivals OpenAI’s o1 in performance but costs only 3% in comparison. This breakthrough not only makes high-performance AI more accessible but also opens up new possibilities for the entire AI industry.
Why is DeepSeek R1 So Important?
1. Perfect Balance of Performance and Cost
- Mathematical ability: Achieved a 79.8% pass rate in AIME 2024 test, surpassing OpenAI o1
- Programming capability: Outperforms 96.3% of human programmers on Codeforces platform
- General knowledge: Achieves 90.8% accuracy in MMLU tests
2. Revolutionary Pricing Strategy
Compared to OpenAI o1’s price of $60 per million tokens, DeepSeek R1 only requires $2.2 for the same task. This means:
- Enterprise annual AI usage costs can be reduced from tens of thousands to hundreds of dollars
- Individual developers can now afford high-performance AI services
Cost Comparison Between DeepSeek R1 and OpenAI O1
| Price per million tokens | Input Price | Output Price |
|---|---|---|
| OpenAI O1 | $30 | $60 |
| DeepSeek R1 | $1.1 | $2.2 |
| Savings Percentage | 96.3% | 96.3% |
Practical Value Analysis
1. A Blessing for Individual Users
Most excitingly, ordinary users can now run AI models approaching o1’s level on their own computers:
- Mac users can deploy through simple steps
- Supports regular computers with less than 32GB memory
- No need to pay expensive API fees
2. Enterprise Applications
For business users, DeepSeek R1 offers several attractive features:
- Significantly reduced code security analysis costs
- Supports private deployment
- MIT license ensures freedom for commercial use
Technical Innovation Highlights
DeepSeek R1 incorporates several important technical innovations:
- Perfect Combination of Open Source and Performance
- Achieved 79.8% pass rate in AIME 2024 test
- Outperforms 96.3% of human programmers on Codeforces platform
- Achieves 90.8% accuracy in MMLU tests
- Flexible Deployment Options
- Supports local deployment and operation
- Compatible with computers having less than 32GB memory
- Supports multiple operating systems including Mac
- Provides private deployment solutions
how ot use DeepSeek R1
To start using DeepSeek R1, follow these steps:
- Online Access
- Visit the official website directly: https://chat.deepseek.com/
- Register for free to start using
- Quick Local Deployment
Download and install Ollama: Visit https://ollama.com/download
Run suitable model version:
# 1.5B version (smallest)
ollama run deepseek-r1:1.5b
# 8B version (recommended for beginners)
ollama run deepseek-r1:8b
# 14B/32B/70B versions (requires stronger GPU)
ollama run deepseek-r1:14b
ollama run deepseek-r1:32b
ollama run deepseek-r1:70bUsing Chatbox client (optional):
- Download and install https://chatboxai.app/
- Select Ollama as model provider in settings
- Set API address to http://127.0.0.1:11434
- Select installed model to start conversation
Reference: https://www.reddit.com/r/ollama/comments/1i6gmgq/got_deepseek_r1_running_locally_full_setup_guide/
Running speed and memory requirements on different GPUs
source:https://www.reddit.com/r/LocalLLaMA/comments/1i69dhz/deepseek_r1_ollama_hardware_benchmark_for_localllm/
Conclusion
The release of DeepSeek R1 not only brings technological breakthroughs but more importantly, it ushers in a new era of AI democratization. It proves that high-performance AI doesn’t have to be expensive proprietary technology but can become a productivity tool accessible to everyone. Whether individual developers or enterprise users, everyone can leverage this open-source model to gain more opportunities in the AI era.