Last week, two Chinese AI companies unveiled models that they say can credibly compete with leading systems from OpenAI and Anthropic. The response was quick and predictable. Markets faltered, analysts declared that Silicon Valley was shaking, and policymakers reached for the familiar language of arms races and wake-up calls.
In one article, The Associated Press he said the Chinese model surprised “the US tech industry”. Bloomberg described it as a “remarkable success” that “led the markets” and sent global stocks tumbling on concerns that it could force US firms to rethink how they spend on data centers, chips, and other AI infrastructure. Business Insider asked if the launch is “The next DeepSeek?”, referring to the Chinese model that blindsided the US AI industry last year. Xprize founder Peter Diamandis even called the release America’s “AI Sputnik moment,” referring to the Soviet satellite launch during the Cold War that spurred significant US investment in its science and space programs. Of course, DeepSeek was also widely described as America’s AI Sputnik moment, a comparison that felt meaningless at the time as DeepSeek seemed to arrive with little warning, challenged existing ideas about the cost of the frontier of AI, and caused an immediate reaction in all sectors of technology and finance.
What’s actually surprising is that the model announcements have been surprising at all. For years, we’ve been warned that China is getting involved in AI. However, the world is shocked when it starts to look like the moment has arrived.
US and Chinese companies train almost all of the world’s most widely used AI models, and six of the top 10 AI tools on OpenRouter’s token usage leaderboard and benchmarks were Chinese. The performance gap has been narrowing for some time, with the latest models from companies like Z.ai and DeepSeek seen as more competitive with high-end offerings from US labs like Anthropic and OpenAI. Chinese models are also much cheaper to use, and reports suggest that US companies are increasingly turning to Chinese equipment as the cost of using domestic suppliers rises.
Beijing is also interested in supporting domestic AI efforts, including encouraging and funding innovation and cracking down on firms that try to ditch their ties to China. Meanwhile, Washington’s AI strategy has often veered between heavy-handed interventions that have left allies questioning America’s credibility and assumptions that markets will see things in the right light. It is a difficult strategy to maintain against a competitor who is ready to concentrate the full power of the empire behind a single technological goal.
Beijing-based startup Moonshot AI, one of China’s leading AI model developers, launched a new flagship model on Friday, saying it outperforms almost every US model, trailing only OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Fable 5. Moonshot also priced the Kimi K3 at R3 million in dollar pressure 3.0 million with a charcely of $3. for GPT-5.6 Sol and $50 for Fable 5. Demand was so strong after launch that Moonshot, the company said, even temporarily suspended new subscriptions after the service became overwhelmed. Most of the answers are focused on this release.
Days later, Chinese tech titan Alibaba followed with a preview of Qwen3.8. It described the new model as “one of the most powerful models[s] available today” and “only the second in Fable 5.” This only adds to the chaos created by Kimi K3.
Most importantly, both companies plan to make their new flagship models publicly available. Both Moonshot and Alibaba say they plan to release their models as an open source, which would allow developers to download, use, and modify the basic values created during AI training that shape its responses. It differs greatly from the closed, proprietary approach to frontier models taken by leading US AI labs, including OpenAI, Anthropic, and Google.
Economics need to be looked at carefully. There is a whole unresolved debate about the extent to which Chinese companies – as American companies suspect – use American models to train their own, which can improve performance at a fraction of the cost. Tokens are not exactly the same between models, and token prices alone give an incomplete picture of how much it costs to run an AI system. A more expensive model may, for example, produce better answers with fewer tokens. Companies also often subsidize inference costs to benefit customers. Cheaper, in other words, doesn’t automatically mean better, or less expensive overall.
Still, it’s possible that Chinese labs could end up producing models that aren’t just cheap substitutes, but systems that can truly match or outperform their US competitors. Even companies that follow the frontier a little can still have a big impact if their models are good enough, easy or cheap to ship, or available on attractive terms. This can have direct consequences for US companies, the broader economy, and national security.
Anthropic and OpenAI are both preparing for what could be $2 billion IPOs, prices that depend in part on the hope that they will dominate the global AI market. China’s talent models challenge that assumption, and may drive away customers, squeeze margins, and weaken the growth projections that underpin that valuation. Given how expensive American AI has become, some US startups are already turning to cheaper Chinese models. There are broader market risks, too, that reach far beyond a handful of AI players. Tech stocks make up a large portion of the US market, and much of that recent growth has been tied to expectations that demand for AI will continue to rise. Companies have piled hundreds of billions of dollars into data centers, chips, power, and other infrastructure relying on the assumption that American firms will continue to dominate. If Chinese labs are able to capture some of that demand, or show models that can be produced and used for less money, investors can definitely question whether those costs are worth it. Given the money involved, any reassessment would have devastating consequences for all of these industries, as well as the millions of people whose savings or pensions are exposed to them.
There are security considerations, too. China’s more open models, even if they follow the US border, can make advanced AI systems available to a wider range of users, especially in cases where US companies limit access or impose strong protections. While the US government wanted Anthropic to restrict access to its latest models, cybersecurity leaders warned that doing so would make it harder for defenders to find and fix vulnerabilities. Those limitations are hard to justify when comparable models are available elsewhere. Organizations denied access to US models may feel forced to rely on Chinese alternatives to protect their networks, or accept greater exposure to attackers who are able to use similar tools. Already, reports are starting to emerge that Kimi K3 has identified and fixed a cyber vulnerability that will not affect OpenAI’s Codex and Anthropic’s Fable due to security reasons. Even the most inefficient models can still be a threat and some just seem to be. In June, China’s Z.ai said its GLM-5.2 model could match Anthropic’s Mythos in cybersecurity tasks, though it trailed in more general tasks.
Since neither model has been fully released, it’s still difficult to independently assess how powerful it is, and companies’ benchmark claims should be treated with caution. However, there has been little public suggestion that companies are distorting their results when it comes to performance.
But the exact level is almost beside the point. Whether Kimi K3 and Qwen3.8 ultimately prove to be among the world’s top five models or the top ten, the broad conclusion remains the same: China’s leading AI companies are now producing systems that can clearly compete with those coming out of top US labs. And they do this regularly enough that each new release no longer needs to be treated as a shock, let alone something simultaneously reinforcing like another “DeepSeek” or “Sputnik moment.” If this really is a race, it’s time to accept that someone else might actually win, or at least come close enough to it.
