OpenAI President Praises Kimi K3 as 'Pretty Good': US-China AI Competition Enters New Phase
On July 22, 2026, OpenAI President Greg Brockman rarely publicly evaluated a Chinese competitor's AI model during a Bloomberg interview. When asked about Moonshot AI's newly released Kimi K3 model, Brockman stated: 'It's a pretty good model. There's no question about it.' But he also added uncertainty about whether the model was distilled from OpenAI's GPT series. This evaluation is significant—it's the first time an OpenAI executive has publicly acknowledged that a Chinese AI model has genuine competitiveness, marking a new phase in US-China AI competition.
Kimi K3 is Moonshot AI's latest model released on July 17, with 2.8 trillion parameters, making it one of the world's largest AI models. According to Moonshot's test data, while Kimi K3 still trails Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol on overall performance, it consistently outperformed Claude Opus 4.8 and GPT-5.5—Anthropic and OpenAI's second-flagship models respectively—on benchmarks including coding and general agents. Bank of America analysts noted in a report: 'Despite persistent hardware/compute capacity constraints in China, K3 demonstrates that pre-training scaling, paired with architectural innovation, can still deliver step-change gains for flagship Chinese models.'
Brockman's concerns about 'distillation' are not without basis. In the AI industry, 'distillation' refers to the technique of using a large model's outputs to train a smaller model. Some analysts suspect that Chinese AI companies may have accessed OpenAI's models through APIs and then used their outputs to accelerate their own model training. If these allegations are true, it would involve serious intellectual property and technology transfer issues. U.S. Treasury Secretary Scott Bessent even threatened sanctions against Chinese AI labs earlier this week, citing 'IP theft' concerns. However, Moonshot AI firmly denied distillation allegations, claiming Kimi K3 is the result of completely independent research and development.
Kimi K3's release triggered a chain reaction in China's AI industry. On the day of the announcement, shares of other Chinese AI companies plummeted. Z.ai, another Chinese AI company that released a new model to much fanfare in June, saw its stock drop 28%. MiniMax Group also fell 16%. Investors worried that Kimi K3's strong performance would squeeze the survival space of other Chinese AI companies. Meanwhile, although Alibaba's stock had risen on news of its partnership with Apple in China, it also dropped 4% after Kimi K3's release. Analysts believe Alibaba's Qwen 'open-source leader' narrative may face some tests.
This event also reflects profound changes in the global AI competitive landscape. Previously, US-China AI competition was mainly a pattern of 'US leading, China catching up.' But Kimi K3's emergence shows that in specific domains, Chinese AI models can already compete with American second-flagship models. More importantly, Moonshot AI plans to conduct an IPO, meaning Chinese AI companies are moving from technology R&D to capital markets, preparing to compete head-on with American giants on the global stage. Brockman's 'pretty good' evaluation may seem bland, but it actually represents OpenAI's formal acknowledgment of Chinese AI strength. This acknowledgment may foreshadow more cooperation in the future—or more intense competition.
🤔 Frequently Asked Questions
Q1: How does Kimi K3 perform?
Kimi K3 has 2.8 trillion parameters. While overall performance trails Claude Fable 5 and GPT-5.6 Sol, it surpasses Claude Opus 4.8 and GPT-5.5 in coding and agent tasks.
Q2: What is AI model 'distillation'?
Distillation is a technique using a large model's outputs to train a smaller model. OpenAI suspects Kimi K3 may have used GPT model outputs for training, but Moonshot AI denied this allegation.
Q3: What does this mean for investors?
Kimi K3's release caused other Chinese AI company stocks to plummet, indicating intensified competition. Investors need to focus on each company's technological differentiation and commercialization capabilities.
Q4: What's the future of US-China AI competition?
Chinese AI models are rapidly closing the gap and can already compete in specific domains. The future may see more cooperation or more intense competition, with geopolitical factors continuing to influence this landscape.
🛠️ Recommended Tools
Summary
OpenAI President's 'pretty good' evaluation of Kimi K3 may seem casual, but it's actually an important signal of changing US-China AI competition dynamics. It shows that Chinese AI models are no longer simple followers but have genuine competitiveness in certain domains. Kimi K3's 2.8 trillion parameter scale, excellent performance in coding and agent tasks, and Moonshot AI's IPO plans all show that Chinese AI is shifting from 'catching up' to 'running alongside' or even 'leading.' However, distillation controversies and geopolitical tensions also add uncertainty to this process. In the future, US-China AI competition will be more complex—involving not just technical competition but multi-dimensional games around intellectual property, national security, and market access. For the global AI industry, this competition is both a challenge and an opportunity that will drive the entire industry forward faster.