TECH

Moonshot Open-Sources Kimi K3 as U.S.-China AI Tensions Intensify

Published: Jul. 29, 2026 12:17 a.m. GMT+8

00:00

00:00/00:00

Listen to this article 1x

The Kimi K3 booth at the 2026 World Artificial Intelligence Conference in Shanghai on July 20, 2026. Photo: VCG

Chinese artificial-intelligence startup Moonshot AI has open-sourced the weights of its flagship Kimi K3 model, even as U.S. scrutiny intensifies over allegations of intellectual-property theft and a broader geopolitical debate over AI safety.

Late on Monday, Moonshot released Kimi K3’s model weights, technical report and supporting infrastructure technologies. The company said open-weight models can lower barriers to access, spur innovation and give users more privacy and control over their data.

You've accessed an article available only to subscribers

Subscribe today for just $.99.

VIEW OPTIONS

Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.

Save an extra $50. Introductory offer for new readers. Subscribe now.

Share this article

Open WeChat and scan the QR code

DIGEST HUB

Digest Hub

Back

Explore the story in 30 seconds

Moonshot open-sourced Kimi K3, lowering barriers and sparking U.S. IP theft allegations.
U.S. threatens sanctions, intensifying open-source AI debate.
Kimi K3 ranks third globally; Moonshot eyes Hong Kong IPO.
Kimi K3 ranks third globally; Moonshot eyes Hong Kong IPO.

AI generated, for reference only

Explore the story in 3 minutes

1. Chinese AI startup Moonshot AI open-sourced the weights, technical report, and support infrastructure for its flagship Kimi K3 model late Monday, stating the move lowers barriers, spurs innovation, and gives users more privacy [para. 1][para. 2]. This escalated the U.S.-China technology dispute, prompting calls from prominent U.S. executives for tighter limits on China’s AI development while simultaneously energizing the industry defense of open-source models [para. 3].

2. Responding to a pro-open-source letter led by Nvidia's Jensen Huang, Anthropic CEO Dario Amodei agreed with much of the letter but argued that open weights do not inherently make security easier [para. 4]. He warned that China could develop models more powerful than those in the U.S. to secure a lasting military advantage or strengthen domestic repression, and raised concerns about misuse for cyberattacks or biological threats [para. 5]. He reiterated calls for tighter chip export restrictions, stronger action against industrial-scale model distillation, and mandatory safety testing for frontier models [para. 5].

3. China's Ministry of Commerce rejected the U.S. allegations of distillation and intellectual property theft as lacking factual and legal basis, labeling the American stance a double standard and an example of "AI hegemony" [para. 6]. Since K3's debut on July 17, tensions rapidly escalated. On July 22, White House advisor Michael Kratsios alleged Moonshot had distilled Anthropic's Fable model, and the following day Treasury Secretary Scott Bessent warned that sanctions or Entity List placement were possible if IP theft was confirmed [para. 7].

4. An open letter led by Nvidia CEO Jensen Huang on July 24 argued that open-source models are vital for security, innovation, and technological autonomy [para. 8]. Initially backed by 25 organizations including Meta, Microsoft, and Dell, the letter advocated for targeted rules over sweeping restrictions [para. 8]. By Tuesday, signatories had grown to 132, including Amazon, Google, and OpenAI, and a separate coalition of nearly 200 Silicon Valley companies also publicly opposed sanctions [para. 9].

5. Kimi K3 features 2.8 trillion parameters and a mixture-of-experts architecture, making it the largest open-weight model, surpassing DeepSeek-V4-Pro [para. 10]. It supports native visual understanding and a 1-million-token context window [para. 10]. Alongside K3, Moonshot released MoonEP, a communications library, and AgentEnv, a sandbox system [para. 11]. The model ranks third globally on Artificial Analysis behind Claude Fable 5 and GPT-5.6 Sol, and first for coding on Code Arena [para. 12]. Strong launch demand forced Moonshot to temporarily suspend new user subscriptions due to computing constraints, which also weighed on the share prices of domestic rivals [para. 12].

6. Against this backdrop, Moonshot is reportedly exploring a Hong Kong IPO within six months and a funding round that could lift its valuation above $30 billion, with annual recurring revenue having reached $300 million in June [para. 13].

AI generated, for reference only

Who’s Who

AI generated, for reference only

What Happened When

July 17, 2026：: Moonshot AI initially debuts the Kimi K3 model.

July 19, 2026：: Bloomberg reports that Moonshot is exploring a Hong Kong initial public offering within the next six months and pursuing a funding round that could lift its valuation above $30 billion, with annual recurring revenue reaching $300 million in June 2026.

July 22, 2026：: White House science and technology adviser Michael Kratsios alleges that Moonshot had distilled Anthropic’s latest Fable model to develop K3.

July 23, 2026：: Treasury Secretary Scott Bessent says the U.S. could consider sanctions or placement on the Entity List if Chinese companies were secretly carrying out large-scale model distillation attacks.

July 24, 2026：: Nvidia founder Jensen Huang leads an open letter arguing that open-source models are important for security, innovation, and technological autonomy, initially backed by 25 organizations including Meta, Microsoft, and Dell.

Late on Monday, July 28, 2026：: Moonshot AI releases Kimi K3’s model weights, technical report, and supporting infrastructure technologies. Also on this date, Anthropic CEO Dario Amodei publishes a statement responding to the open letter by Huang, and a China’s Ministry of Commerce spokesperson rejects U.S. threats.

AI generated, for reference only