Chinese AI startup Moonshot’s latest K3 large model is shaking up the global tech market. Analysts point out that the launch of this brand new model has upended the prevailing notion that the United States still leads China in the field of AI.
On Friday (July 17), Moonshot released the new K3 model. In multiple benchmark tests, this model not only outperformed competitors but its capabilities also rival those of the top products from American giants OpenAI and Anthropic. The news swiftly triggered volatility in global tech stocks.
According to Bloomberg, previously, the industry—including China’s top AI leaders and leading US companies—generally believed that China still lagged significantly behind the US in cutting-edge AI model development, with the gap even widening. Just this week, an anonymous executive from Anthropic assessed that their company’s technology is around 6 to 12 months ahead of Chinese competitors.
However, the comprehensive abilities shown by K3 are breaking this assumption. University of California, Berkeley computer science professor Ion Stoica told Bloomberg: “We used to say that open source models from China and the like lagged frontier models by about six to nine months. But now, that gap may have shortened to two to three months.”
However, the comprehensive abilities shown by K3 are breaking this assumption. University of California, Berkeley computer science professor Ion Stoica told Bloomberg: “We used to say that open source models from China and the like lagged frontier models by about six to nine months. But now, that gap may have shortened to two to three months.”
The performance leap of K3 has sparked a wave of tech stock sell-offs similar to last year’s “DeepSeek moment.” This has once again triggered market concerns: can Silicon Valley’s massive investments in computing power yield the expected returns? Has the US’s absolute leadership in AI been shaken?
Bloomberg reports that Moonshot’s progress could complicate efforts by US regulators to defend domestic superiority. Under regulatory pressure that requires prior review, both OpenAI and Anthropic have slowed their product release schedules during the Trump administration. Now, this slowdown is likely to give China more time to catch up, thereby threatening the White House’s ambition to maintain AI supremacy.
As customers become increasingly sensitive to soaring AI spending, some developers have started to distribute certain tasks to low-cost options, including Chinese models.
Bloomberg reports that Moonshot’s progress could complicate efforts by US regulators to defend domestic superiority. Under regulatory pressure that requires prior review, both OpenAI and Anthropic have slowed their product release schedules during the Trump administration. Now, this slowdown is likely to give China more time to catch up, thereby threatening the White House’s ambition to maintain AI supremacy.
As customers become increasingly sensitive to soaring AI spending, some developers have started to distribute certain tasks to low-cost options, including Chinese models.
However, Moonshot also acknowledged the limitations of its new product. An official blog post admits that although K3 is highly competitive overall, there is still a notable gap in user experience compared to Western models.
Unlike DeepSeek, which focuses on ultra-low prices, the K3 comes with a hefty price tag due to its 2.8 trillion parameters, even comparable to Anthropic’s mid-range models. Bloomberg believes Moonshot seems to be betting that K3’s outstanding performance in programming and high-value agent tasks will support its premium pricing.
As of Friday afternoon, K3 had reached the top of the Arena leaderboard, which evaluates AI model web development tasks. Stoica pointed out that early results show that K3 is already “in the same ballpark” as Anthropic’s Fable and OpenAI’s GPT-5.6 Sol in the programming field.