OpenAI president says it's too early to judge whether Kimi 'distilled' its models

OpenAI co-founder and President Greg Brockman said he is not certain whether Kimi K3, an AI model from Chinese startup Moonshot AI, was built by distilling OpenAI's own models.

In an interview with Bloomberg published Wednesday, Brockman said it was "still too early" to determine whether Kimi K3 had engaged in distillation. He also called Kimi K3 "quite a good model, without question."

Brockman said he had seen estimates that China lags the United States by about four months in AI model development, and said OpenAI still holds a "massive advantage" over other AI companies thanks to its early investment in computing resources and years of research.

Brockman also pushed back on the notion that open-weight models are free, calling it a "misconception." He said such models "are also large models running on a lot of compute, and compute is expensive."

Kimi K3 is a recently released AI model from Moonshot AI that has been assessed as outperforming Anthropic's Claude Opus 4.8 on some benchmarks. Following DeepSeek last year, Kimi's emergence has shaken the industry and demonstrated the rapid growth of Chinese AI technology, prompting US voices to allege that Chinese companies have distilled American AI models — in effect stealing intellectual property.

Michael Kratsios, director of the White House Office of Science and Technology Policy, posted on X that day claiming he had received information that "Moonshot AI distilled Anthropic's (Claude) Fable." He alleged that Moonshot AI had "developed a sophisticated internal platform capable of performing large-scale distillation of US models" and had "rapidly diversified its approach to avoid detection."

Treasury Secretary Scott Bessent, in a Fox News interview on Wednesday, said the US would investigate allegations that Chinese companies distilled American AI models and would impose sanctions on any firms found to have done so. "We are finding 'watermarks' of US large language models in many Chinese models. This is unacceptable," he said, adding that the administration would "look into this in the coming days or weeks."

Bessent said that if Chinese AI companies had distilled US models as alleged, it would constitute "intellectual property (IP) theft," and added: "We have the ability to sanction them for this theft."