A public conversation about the positions of Chinese and American large models was taken up for the first time by a core figure from OpenAI. On July 23, according to Bloomberg, Greg Brockman, president and co-founder of OpenAI, gave a rare and straightforward evaluation of Kimi K3, released by Moonshot AI in the previous week: this model is indeed very good, there's no doubt about it.

The weight of this statement lies in the fact that it comes from a key figure at a company long seen as a benchmark for AI in the United States. In the past, people generally assumed that the US had a significant lead over China in foundational models. However, the release of Kimi K3 is challenging this perception—Moonshot AI claims that its open-weight model has surpassed all competitors except Anthropic Claude Fable5 and OpenAI GPT-5.6. The market reaction followed quickly.

However, while acknowledging the strengths, Brockman left an open question. He said it was too early to determine whether Moonshot AI obtained the outputs of OpenAI models through "distillation" to train Kimi K3, and emphasized that OpenAI has always been monitoring such activities. This statement once again brought the technical boundary debate between open-weight models and native cutting-edge models back into the spotlight.

More intriguing was his quantitative judgment of the gap. Brockman cited some estimates stating that China may only be behind the US by about four months in AI model development, with other experts even suggesting the gap is smaller. As low-cost Chinese models spread rapidly and the capability gap continues to narrow, closed-source companies like OpenAI and Anthropic, which rely on subscriptions and API revenue, face more uncertainties—companies that are currently preparing for their IPOs and working to increase customer revenue.

Facing the approaching competitors, Brockman's confidence comes from two points: the substantial computing resources invested early on and years of accumulated AI research experience. Based on this, he believes that OpenAI still maintains a significant advantage over its competitors and stated that the company has been deeply studying how to build more efficient models that can precisely accomplish set goals.

He also took the opportunity to clarify a common misconception: open-weight models like Kimi are not free products. He reminded that these models are large in scale and require a lot of expensive computing power to run. The real competition is not about open source or not, but about who can build the most efficient, smartest, and best value-for-money models in specific tasks. When competitors approach the capability level with lower prices, what OpenAI needs to protect might no longer just be its position on model rankings, but an increasingly hard-to-define value advantage.