DEEPSEEK-V3 IS NOW THE BEST OPEN SOURCE AI MODEL

Summary from the TLDR newsletter

DeepSeek, a Chinese AI research lab, has released DeepSeek-V3, a Mixture-of-Experts model that features a total of 671B parameters. The model was trained on 14.8 trillion tokens. It has been released on GitHub along with a detailed paper outlining its capabilities. The model outperformed Meta's flagship Llama 3.1 405B parameter model and many other closed-source models on benchmark scores.