Sat Jan 25 05:20:00 UTC 2025: ## Chinese AI Model DeepSeek-R1 Rivals OpenAI’s o1 at a Fraction of the Cost

**Hangzhou, China** – A new large language model (LLM) developed by the Chinese company DeepSeek is generating significant excitement within the scientific community. DeepSeek-R1, released on January 20th, offers comparable performance to OpenAI’s highly-regarded o1 model in chemistry, mathematics, and coding tasks, but at a drastically lower cost and with greater accessibility.

Unlike OpenAI’s models, which are largely considered “black boxes,” DeepSeek-R1 is released under an MIT license, allowing researchers to study and build upon the algorithm. While not fully open-source (training data remains unavailable), this “open-weight” approach is unprecedented for such a powerful model.

Researchers have praised the model’s affordability. Running DeepSeek-R1 is estimated to cost approximately one-thirtieth of the expense of using o1. DeepSeek has also created smaller versions of the model, making it accessible to researchers with limited computing resources. One researcher reported that an experiment costing over £300 with o1 cost less than $10 with R1.

The development of DeepSeek-R1 is particularly noteworthy given US export controls that restrict Chinese access to high-end AI processing chips. Experts suggest that DeepSeek’s success demonstrates the importance of efficient resource utilization over sheer computing power.

This achievement has led some to question the previously perceived US lead in AI development. The emergence of DeepSeek-R1, following the release of their cost-effective V3 chatbot, suggests a narrowing gap in technological capabilities between the US and China. Calls for international collaboration in AI development are growing in response to this rapid advancement. DeepSeek-R1 represents a significant step forward for open and affordable AI research.

Read More