ChainThink reports that on July 18, Dean W. Ball, Head of Strategy at OpenAI, stated that Kimi K3 is a highly capable model whose performance in agent programming is nearing that of the best publicly available models expected in the first quarter of 2026, and these capabilities cannot be simply explained by distillation.
Ball believes that the pressure from China's open-source models goes beyond just price competition.
If open models become sufficiently powerful, developers' willingness to continuously pay for proprietary models may decline, affecting model vendors' profits and investor expectations, thereby weakening U.S. companies' incentive to invest hundreds of billions of dollars in training the next generation of models.
He also stated that open weights would accelerate technological diffusion but may compress commercial returns from training state-of-the-art models, making model development more reliant on subsidies from other businesses or government funding, similar to public infrastructure such as power grids and roads.
Ball expects that the U.S. government may, in the future, restrict Chinese models from entering U.S. enterprises—particularly regulated industries such as banks—through measures such as backdoor alerts, data security concerns, and compliance risks;
He believes that even if the evidence is insufficient, creating sufficient uncertainty could influence corporate adoption.