Chinese AI models are quietly taking over the top of the download charts in the US, and even big tech companies are switching to them

[Dataset Cheetah]

The top five most popular models on OpenRouter, a platform tracking usage data across AI models, were all Chinese as of this week, according to Associated Press reporting, and the trend goes well beyond raw download numbers. Mozilla's chief technology officer Raffi Krikorian switched to Moonshot's Kimi K3 for many of his daily tasks within days of its release, saying it just seems snappier than Anthropic's more expensive Claude Fable, and cryptocurrency exchange Coinbase confirmed it is switching to Chinese AI models specifically to cut costs. Kimi K3 saw more than 930,000 downloads in the week after its July release, a 200% jump from the week before, with roughly 86,000 of those downloads coming from the US alone, up 387% week over week

The appeal comes down to cost rather than raw capability. Curt Meinhold, a North Carolina based tech executive, said most users, well over 90%, simply don't need Anthropic's Mythos or Fable, they need something good enough, paying a handful of cents per million output tokens for a Chinese model versus 30 to 50 dollars for a premium American one. Goldman Sachs described Chinese models as reaching a critical stage for wider adoption, particularly as agentic AI use, where models autonomously complete multistep tasks, compounds those per token cost differences dramatically

The Trump administration accused Moonshot this week of using covert but not necessarily illegal methods to build K3 off the back of Anthropic's Fable model, a distillation claim Beijing rejects as groundless. Meanwhile, a group of American tech firms including Microsoft, Meta and Nvidia published an open letter this same week backing open AI models generally, and some analysts noted that US export controls that took Anthropic's own Fable and Mythos offline for over two weeks earlier this year directly created an opening for Chinese competitor Z.ai to release its GLM-5.2 model into that exact gap. Most Chinese models remain open source, while frontier American models from Anthropic and OpenAI stay closed, and Mozilla's Krikorian summarized the shift bluntly, the open frontier is becoming increasingly Chinese-built

Cheaper, open, intelligent: Chinese AI models gain ground, as they make inroads in USabcnews.com

[Chris81]

The point about US export controls directly creating an opening for a Chinese competitor is the most important, underappreciated detail in the whole piece, restricting one country's model can hand market share straight to another

[Gareth84]

Coinbase specifically switching for cost reasons is a real signal, that's a serious financial company making a purely economic calculation rather than an ideological one about openness

[Southern Joanne]

The good enough framing from that tech executive captures something important, most actual daily use cases genuinely don't need frontier level capability, they need cheap and reliable

[BookerT]

Worth remembering these are self reported download numbers and switching decisions from individuals, not necessarily proof Chinese models have closed the capability gap entirely, just that the cost gap is winning a lot of real world decisions

[BankHolidayBlues]

The distillation accusation against Moonshot getting no supporting evidence again here is the same pattern from when this story first broke, a serious claim with nothing concrete attached to it yet

[NightHarbour]

Open source Chinese models versus closed source American frontier models is going to be one of the defining structural differences in how this whole industry develops over the next few years

[Phoenix56]

Genuinely useful reporting that treats this as an economic story about cost and access rather than pure geopolitical alarm, that framing feels more accurate to what's actually driving individual adoption decisions

[Vector14]

This being framed as no longer simply America versus China but Chinese labs also competing hard against each other is an important nuance, the competitive pressure domestically in China is intense too

[RightAbout24]

930,000 downloads in one week for K3 is an enormous number for any software release, that alone shows real product market fit beyond just curiosity

[Northern Squid]

The open letter from Microsoft, Meta and Nvidia landing the same week as this story is clearly not a coincidence, this whole open model policy fight is happening in real time right now