<ul><li>Chinese AI start-up DeepSeek has released two open-source models - V3 large language model and R1 reasoning model, which achieved performance on par with major Western competitors at a fraction of the cost.</li><li>DeepSeek released eight open-source projects, disclosing details of the techniques used to optimize compute, communications, and storage for model training.</li><li>The open-source approach of DeepSeek contrasts with the closed-source model of Microsoft-backed OpenAI, which recently launched GPT-4.5 at a higher cost compared to DeepSeek's models.</li><li>DeepSeek reduced prices for access to its models and resumed account top-ups for developers after temporarily suspending them due to high demand.</li></ul>

DeepSeek Open-Sources AI Model Training Details

Discover more