OpenAI president says Kimi K3 ‘pretty good’, unsure if distilled from GPT models

OpenAI president Greg Brockman acknowledged that Moonshot AI had developed a competitive new artificial intelligence model and said he was unsure whether the Chinese firm had piggybacked on the ChatGPT maker’s technology.

“It’s a pretty good model,” Brockman said in an interview on July 21, referring to the release days earlier of Moonshot’s more advanced Kimi K3. “There’s no question about it.”

The OpenAI co-founder said it is “too early” to tell whether Moonshot may have illicitly extracted results from OpenAI’s models to improve its own through a process known as distillation.

“This is something we do always monitor for,” he said.

On July 22, a White House official accused Moonshot of improperly using US AI models to build its own. OpenAI rival Anthropic PBC has also previously accused Moonshot of distilling its models. Moonshot has not responded to those allegations.

Last week, Moonshot released Kimi K3, a more advanced open-weight model that it said outperforms all rivals except for Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 on overall capability. The announcement rattled markets and upended long-held assumptions about the extent of the US lead over China in the global AI race.

Brockman said he has seen estimates suggesting China could be as little as four months behind the US in model development. Other AI experts have suggested it may be even less.

The proliferation of more affordable Chinese models and the narrowing gap in capabilities inject new uncertainty into the business models for OpenAI and Anthropic. Both firms are working to bolster revenue from customers ahead of their public market debuts.

Brockman stressed that OpenAI continues to have “a huge advantage” over rivals thanks to its early investments in computing resources and years of AI research. “We focused on, for many years, doing lots of research into how do you build efficient models that can achieve a goal,” he said.

He also said there is a “misconception” that open-weight models such as Kimi are free. “These are still big models that run on lots of compute, and compute is expensive,” he said.

“The competition really will be about, ‘How do you build the most efficient, intelligent models that get the best task price performance?’”

As Chinese models improve, some have speculated the Trump administration could move to ban access to them. In a statement, OpenAI said open models are important for “democratising access to AI” and added that “advances in Chinese open-weight models are not an argument against openness”.

Brockman, who has previously donated to President Donald Trump’s super political action committee, declined to wade into the debate. In general, though, he emphasised that it is important “America continues to lead” in AI, and expressed optimism that it will.

“You’re going to continue to see very, very special models continue to come from the leading labs,” he said. BLOOMBERG