Greg Brockman, the president of OpenAI, described Moonshot AI’s Kimi K3 model as “pretty good” during a Bloomberg interview on Tuesday. The comment marked one of the few public acknowledgments from a leading U.S. lab that a Chinese competitor has produced a system capable of matching top American models on several benchmarks.

Brockman stopped short of confirming whether Moonshot distilled K3 from OpenAI’s own outputs, saying the evidence is still too early to evaluate. His measured tone contrasted sharply with recent statements from the White House, where officials leveled direct accusations of intellectual‑property theft against the Beijing‑based startup.

White House and Treasury Accusations

Office of Science and Technology Policy director Michael Kratsios alleged that Moonshot accessed Nvidia chips that are prohibited for export to China, routing them through intermediaries in Thailand. He also claimed the company carried out large‑scale distillation of U.S. AI models, a practice OpenAI and other firms have warned about for months.

On the same day, Treasury Secretary Scott Bessent announced that the administration would scrutinize Chinese open‑source models for stolen IP and warned that sanctions could follow within weeks. The Treasury’s stance underscores a shift from viewing the dispute as a commercial rivalry to treating it as a national‑security issue.

Industry Reactions

Nvidia chief executive Jensen Huang pushed back against the alarm, calling Chinese open models “excellent” and urging the U.S. government to allow American firms to leverage them. Huang’s remarks reflect the broader industry debate about whether open‑weight models truly democratize AI or merely accelerate commoditization.

The distillation charge has been building. In February, Anthropic accused Moonshot, DeepSeek and MiniMax of running roughly 16 million exchanges through about 24 000 fraudulent accounts to siphon knowledge from its Claude models. OpenAI has made similar complaints, and those allegations now appear to be gaining political traction.

During the interview, Brockman emphasized the United States’ “huge advantage” stemming from billions of dollars invested in compute infrastructure. He argued that open‑weight models like K3 are not free in practice because scaling them still requires expensive hardware, a point that undercuts the narrative that Chinese labs can simply give away frontier‑level AI for little cost.

Brockman estimated China lags the United States by roughly four months in overall model development, a gap he suggests could widen as U.S. firms continue to pour resources into next‑generation hardware and training pipelines.

Kimi K3, released on July 16, boasts nearly three trillion parameters and quickly rose to the top of Arena’s frontend‑coding leaderboard, outpacing both Anthropic and OpenAI flagship offerings within 24 hours. Moonshot priced the model at a fraction of what U.S. labs charge, intensifying the push toward a more affordable model layer that investors have warned about since DeepSeek’s price cuts earlier this year.

When asked whether Chinese models should be banned, Brockman declined to wade into the policy debate. Instead, he reaffirmed OpenAI’s belief that open models are essential for “democratizing access to AI,” a stance that aligns more closely with Nvidia’s CEO than with the administration’s hard‑line approach.

Beyond the technical debate, Brockman’s political donation to a Trump‑aligned super PAC surfaced in the interview, adding another layer to the complex interplay of technology, policy and partisanship that surrounds the AI race.

Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.