The US is approximately 6–8 months ahead of China in frontier AI model capability despite chip export controls.

Chinese frontier AI models are roughly 6-9 months behind US models due to distillation.

“The Chinese are very capable, very smart, and also definitely distilling these models to sort of stay about 6 to9 months behind.”

The US remains approximately 6 months ahead of China in frontier AI capabilities; China has not caught up.

“I don't believe that China has caught up. I still think we are 6 months ahead.”

The US is approximately eight months ahead of China in frontier AI model capability, despite chip export controls.

“We're kind of eight months ahead in terms of where the frontier model is like our frontier model versus their frontier model.”

Leading Chinese LLMs are approximately 6 months behind US models in capability.

“The leading LLMs in China are probably 6 months behind where we are in the US in terms of the capability of their models.”

Training a model close to frontier capability is not technically hard today, and many actors are achieving this not just via distillation.

“It's not hard to train a model that is close to frontier capability.”

Other AI developers are only 3 to 6 months, and certainly less than a year, behind OpenAI in capabilities.

“Other developers are not that far behind probably not more than 3 to 6 months certainly not more than a year behind open AAI in their capabilities.”

The US open-source AI model Thinking Machines was distilled from the Chinese model Kimi K2.5, meaning American open-source innovation is now building on Chinese foundations.

“It was distilled off a Chinese model, Kim K 2.5.”

China's Kimi K3 model has reached the frontier, and the US may only have months of lead over China in AI.

“By the way, Kimmy K3 just came out and people are saying it's it's now right up there. It's very close to the frontier. We may have months on China if that.”

Within a few years, the lead time at the AI frontier will shrink from months to weeks or days.

“In a few years it'll be more like weeks or days.”

DeepSeek achieved GPT-4-level performance at 1/600th the cost of GPT-4.

“Deep Seek, for example, on GPT-4 was 1/600 the cost.”