DeepSeek's V3 AI model gets a major upgrade - here's what's new

Elyse Betters Picaro / ZDNET Not that it ever left, but it appears Chinese AI startup DeepSeek is back in the news -- this time with an updated version of its V3 model, released in December. Also: The best AI chatbots On Tuesday, the company officially announced V3-0324, named after its release month and day. A day earlier, people noticed DeepSeek had uploaded the new model to HuggingFace, but with little additional information. What's new in DeepSeek's V3-0324 model? Like R1 -- DeepSeek's top-performing model released in January and an OpenAI competitor -- the new version is open source (in that its weights are public, not its actual code) under an MIT license. In a post on X, DeepSeek noted that the update shows better coding skills for web development and a "major boost in reasoning performance," but it still recommends it be used for less complex reasoning tasks. R1 remains the lab's top reasoning model, ranking in fourth place on the Chatbot Arena. Also: What is DeepSeek AI ... Read full article.