The new open-source models have closed the gap with other state-of-the-art models in coding performance, but its authors say it trails them in “reasoning” performance by three to six months.
DeepSeek releases new V4 series models highlighting efficiency and long context