New Anthropic and OpenAI models widen US lead over China in AI

In short: New AI models from Anthropic and OpenAI are helping the US pull further ahead of China, especially in security and complex reasoning.

What's going on

The New York Times reports that the latest models from Anthropic and OpenAI have extended the United States’ lead in AI. Two recent examples are Anthropic’s Claude Mythos and OpenAI’s GPT-5.5.

One area where the gap shows up is cybersecurity. Claude Mythos is described as being very fast at finding weak spots in software systems. You can think of it like an expert locksmith that can quickly spot and use flaws in many different locks, which can help defenders fix problems faster, but can also help attackers.

Another area is “reasoning,” meaning the ability to solve multi step problems and follow complex instructions. GPT-5.5 is reported to be a clear jump from earlier versions on these kinds of tasks. Reports cited by the Times say the US lead is roughly months, and could grow if US companies keep improving models quickly.

China is pushing hard to catch up. DeepSeek released an open source model called V4, which is offered publicly for others to use and build on, like sharing a recipe instead of selling the meal. DeepSeek has suggested its model is closer to older top US systems, but still behind the newest releases.

What to watch

The rivalry is also turning into a security and policy fight. US AI companies have accused some Chinese groups of trying to copy their models by collecting large numbers of interactions, similar to studying an answer key at scale. US officials are also looking at tighter controls on advanced computer chips, since those chips are a key ingredient for training top AI systems.

Source: NYTimes