appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Coco Feng

Minxiao Chang

Xinmei Shen

China’s Moonshot AI has made its latest artificial intelligence model Kimi K3 available for public download, as its open-source strategy and more support for non-Nvidia ecosystems trigger heated debates across Silicon Valley.
On Monday, the Chinese AI start-up not only released the model’s “weights” – the underlying parameters that encode its intelligence – but also made available key infrastructure, including tools to improve efficiency and stability.
The move has allowed developers around the world, including in the United States, to use, copy, modify and distribute the software – the largest open-weight AI model so far with 2.8 trillion parameters.
The company has not disclosed the AI chips it used for training, but domestic hardware options are quickly becoming available.
The cloud unit of Alibaba Group Holding said its supernode powered by proprietary Zhenwu M890 chips was compatible with Kimi K3, while a computing platform powered by Huawei Technologies’ Ascend 910C also completed adaptation. Alibaba owns the South China Morning Post.
Moonshot itself is diversifying beyond the Nvidia ecosystem.

According to the company, K3 matched Claude Fable 5 and outperformed Claude Opus 4.8 – both developed by US-based Anthropic – in rewriting and optimising graphics processing unit (GPU) kernels. These small programs successfully executed large amounts of parallel tasks for both an Nvidia Hopper GPU and an unnamed alternative chip, Moonshot said.
Kimi K3 also developed a GPU compiler from scratch and independently designed an inference chip prototype within 48 hours, Moonshot said. A compiler is a software tool that translates code into a language understood by the chips.
The release comes as Chinese open-weight models gain wider adoption overseas, raising concerns in Washington over China’s accelerating AI progress.
“Over the past year, an increasing number of US companies have built production applications on Chinese open-weight models, citing the lower cost of inference for using these models on their own infrastructure or through cloud providers hosting the Chinese models”, said Paul Triolo, a partner and technology policy lead at DGA-Albright Stonebridge Group.
The model’s global reception was immediate.
On Hugging Face, a platform widely used for open-source AI collaboration, Kimi K3 attracted more than 4,000 likes within 30 minutes of public release, the fastest growth so far, said platform CEO Clement Delangue on X on Monday.
The overwhelming interest in Kimi K3 “has driven a lot of concern with the US government and Silicon Valley about what to do about open-weight models in general”, said Triolo, adding that Washington has been considering some type of ban on the use of Chinese open-source models but that many US companies oppose such a plan.
Since Kimi K3’s launch earlier this month, the American tech community has increasingly expressed its support for open-source.
Some 80 companies including Nvidia, Microsoft and Meta Platforms signed a letter last week opposing any ban on open-source models, while Nvidia initiated the Open Secure AI Alliance on Monday, a cybersecurity-focused consortium for open AI tools with dozens of members.
While Anthropic was not involved in either, its CEO Dario Amodei on Monday said the company “has never advocated for a ban on open-weights models”.
Chinese state media has voiced support for open-source AI. Yuyuantantian, a social media account affiliated with China’s state broadcaster, said on Monday that the support for openness did not mean “advocating for the unconditional proliferation of all [AI] capabilities” and that there is no “perfect technological path”.

However, Moonshot’s more restrictive open-source licence for commercial users has drawn criticism. The start-up requires a paid arrangement if any firm uses its model as a service and generates revenues of more than US$20 million a year.
Matt White, global CTO of the Linux Foundation, expressed his disappointment over the licence in a post on X, which he said was “the early technique for US open models” and “sets the industry back”.
Not all agree. “Moonshot is just pursuing a pragmatic and well thought out middle ground between fully closed APIs and unrestricted open source”, said Triolo. “Unlike Meta or Google, Moonshot must fund future model training, inference infrastructure, and GPU capacity without the benefit of massive adjacent businesses that generate major cash flows”.
Nevertheless, deploying an open-weight model does not mean free of charge, as compute costs money.
“The intelligence of such complexity becomes unattainable for most firms due to the memory burden, which existing hardware configurations [for most people] are unable to handle,” said Eugene Cheah, chief executive and co-founder of AI inference platform Featherless.ai.
Additional reporting by Vincent Chow


China’s intelligence chief warns of risks from AI as ‘new arena for strategic rivalry’

Hong Kong cements superconnector role with wave of belt and road deals

Hong Kong aims to scale its tech ecosystem and nurture more unicorns: HKSTP CEO

Trump downplays need to check AI development, says he is unwilling to cede edge to China

Artificial intelligence (AI)

Moonshot’s Kimi K3 triggers Silicon Valley debate over bans on Chinese open source models

A Kimi logo is pictured at the Moonshot AI booth during the World Artificial Intelligence Conference in Shanghai on July 18, 2026. Photo: AFP

K3 claimed to match Anthropic’s Claude Fable 5 and outperform Claude Opus 4.8 in rewriting and optimising GPU kernels. Photo: Reuters

Nvidia initiated the Open Secure AI Alliance, a cybersecurity-focused consortium for open AI tools with dozens of members. Photo: dpa

China’s Kimi K3 ‘significantly below’ US rivals in hacking power, study shows

The Moonshot booth promoting its Kimi K3 AI model at the World Artificial Intelligence Conference (WAIC) in Shanghai, July 17, 2026. Photo: Reuters

The Kimi app icon is seen on a mobile phone screen in Beijing on July 17, 2026. Photo: AFP

Global AI experts push back on US ‘distillation’ claims against Moonshot’s Kimi K3 model

A Kimi logo is pictured at the Moonshot AI booth at the World Artificial Intelligence Conference (WAIC) in Shanghai on July 18, 2026. Photo: AFP

The Anthropic logo, a keyboard and a robotic hand are seen in this illustration taken June 5, 2026. Photo: Reuters

Tech

News

China Future Tech

Chinese open-weight models are gaining wider adoption overseas, raising concerns in Washington over China’s accelerating AI progress


Chinese open-weight models are gaining wider adoption overseas, raising concerns in Washington over China’s accelerating AI progress.


Moonshot’s Kimi K3 triggers Silicon Valley debate over bans on Chinese open source models

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Chinese open-weight models are gaining wider adoption overseas, raising concerns in Washington over China’s accelerating AI progress

Chinese open-weight models are gaining wider adoption overseas, raising concerns in Washington over China’s accelerating AI progress