The Download: DeepSeek’s latest AI breakthrough, and the race to build world models

Chinese AI firm DeepSeek released a preview of V4, its long-awaited new flagship model, on Friday. The model can process much longer prompts than its predecessor thanks to a new design that handles large amounts of text more efficiently. While remaining open source, V4's performance matches leading closed-source rivals from Anthropic, OpenAI, and Google. This release marks DeepSeek's first model optimized for Huawei's Ascend chips, a key test of China's ability to reduce dependence on Nvidia hardware.

V4's launch comes amid growing interest in world models—AI systems that understand physical reality, not just digital text. Researchers like Stanford's Fei-Fei Li and AMI Labs' Yann LeCun argue world models can overcome LLM limitations and unlock robotics breakthroughs. Meanwhile, China blocked Meta's $2B acquisition of AI startup Manus on national security grounds, and Google is investing up to $40B in Anthropic at a $350B valuation, highlighting the escalating AI arms race.