appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Ben Jiang

Minxiao Chang

Chinese start-up Moonshot AI has launched the world’s largest open-source artificial intelligence model, claiming to outperform leading US systems from Anthropic and OpenAI in some capabilities, as domestic developers race to challenge American dominance.
Beijing-based Moonshot said its Kimi K3, unveiled late on Thursday, had achieved “open frontier intelligence” with 2.8 trillion parameters, which was significantly larger than previous open models from Chinese competitors, including DeepSeek’s 1.6 trillion-parameter V4 Pro or Zhipu AI’s 744 billion GLM 5 series, making it the largest open-source model so far.
Parameters, a machine-learning term, measure the complexity of AI systems during training, with a higher figure generally indicating a stronger model.
K3 was designed for long-horizon coding, knowledge work and reasoning tasks, the company said in a blog post.
While acknowledging that K3’s “overall performance still trails the most powerful proprietary models,” the company said the model had demonstrated frontier-level performance across a range of evaluations, “consistently outperforming other tested models”, including OpenAI’s GPT-5.5, Anthropic’s Claude Opus 4.8, and rival Zhipu AI’s latest GLM-5.2.

K3 outperformed Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol – two of the most advanced AI systems from US labs – on a few evaluations, including Program Bench and SWE Marathon, according to self-reported benchmark results by the company.
The two benchmarks were each used to evaluate AI’s capability in constructing software projects from scratch to finish and in ultra-long horizon software engineering, demonstrating an AI system’s real-world use.
K3 was also ranked the top-performing model in the Frontend Code Arena, surpassing Claude Fable 5 across five domains – Brand and Marketing, Reference-based Design, Data and Analytics, Consumer Product, Simulations and Content Creation Tools.
It only trailed Fable 5 in Gaming, according to Arena.AI, formerly known as LMArena, an AI model evaluation platform originally launched by researchers at the University of California, Berkeley.
Frontend Code Arena is a benchmark designed by Arena. AI to evaluate AI coding models specifically for building user interface and frontend applications in real-world use scenarios.
In terms of overall intelligence, independent market consultancy Artificial Analysis said Kimi K3 scored 57 on its Intelligence Index, and was comparable to Opus 4.8 and GPT-5.5, Anthropic and OpenAI’s second-best models, respectively. However, it was still behind Fable 5 and GPT-5.6 Sol in terms of overall performance, according to the consultancy.
The Intelligence Index evaluates frontier AI models across real-world professional tasks, including knowledge work, coding, scientific reasoning, complex problem-solving, general intelligence and domain-specific expertise such as banking and research.
With the performance jump, Artificial Analysis also pointed to the increased pricing for Moonshot’s new model.
At US$15 for 1 million output tokens, Kimi K3’s cost per task of US$0.94 was just under GPT-5.6 sol’s US$1.04 and half of Claude Opus 4.8’s US$1.80, though much higher than its Chinese open-weight peers, according to the consultancy.
Moonshot’s latest model comes one month after its crosstown rival Zhipu AI launched GLM-5.2 to global acclaim with enhanced coding capabilities, underscoring an accelerated push among Chinese companies to close the gap with leading US models.
Kimi K3 will initially be available through Moonshot’s Kimi chatbot, Kimi Work desktop application, Kimi Code developer tool, and API platform, with plans to release the model’s full weights on July 27.

Amid a wider tech stock slump this week, Zhipu AI’s Hong Kong-listed shares fell as much as 25 per cent by Friday midday following the Kimi K3 release, as investors reassessed the competitive landscape among China’s leading AI model developers amid a rapid acceleration in frontier models.
The performance of Moonshot’s new model was a “breakthrough [that was] unexpected and drives sector innovation amid fast model iterations,” a team of Jeffries analysts including Thomas Chong wrote in a research note on Friday.
The analysts also noted the cost advantage of Chinese models.
The K3 release came just hours before the opening of the World Artificial Intelligence Conference in Shanghai, where Chinese President Xi Jinping delivered his first-ever keynote at the annual event.
Xi said China was committed to “encouraging open source, openness, collaboration and sharing”, adding that the open approach could boost technological innovation, industrial development and applications so that “all sectors and businesses can benefit from AI”.
Additional reporting by Xinmei Shen


Philippine tribe vows to stand ground against US-led Pax Silica: ‘listen to us’

As the US weighs restrictions on cloud computing, how will China’s AI sector adapt?

Chinese AI chipmaker Hygon plots expansion from data centres to robotics

For China’s hi-tech IPOs, it’s Hong Kong vs the mainland – but how much does it matter?

Artificial intelligence (AI)

Moonshot AI unveils world’s largest open-source AI model as China narrows gap with US rivals

Moonshot AI said its Kimi K3 achieved ‘open frontier intelligence' with 2.8 trillion parameters. Photo: Shutterstock Images

Moonshot’s latest model comes one month after its crosstown rival Zhipu AI launched GLM-5.2 to global acclaim. Photo: Shutterstock Images

The K3 release came just hours before the opening of the World Artificial Intelligence Conference in Shanghai. Photo: Xinhua

As US AI costs soar, global businesses pivot to China’s low-cost, open-weight models

The daily token volume of Zhipu’s GLM-5.2 has surged 50-fold on Vercel since mid-June, according to the San Francisco-based cloud platform for AI web development. Photo: Shutterstock Images

US hardens AI stance on China as Anthropic calls for extending lead

Artificial intelligence has become a central battleground in US-China technological competition. Photo: Shutterstock

Anthropic has called for tighter US export controls on advanced AI technologies. Photo: Reuters

Alibaba is one of China’s leading developers of artificial intelligence models. Photo: Reuters

Sam Altman signals OpenAI price war as rivalry with Anthropic, China heats up

OpenAI CEO Sam Altman speaks at an event in Tokyo, Japan, last year. Photo: Reuters

Tech

Tech Trends

Kimi K3 is designed for long-horizon coding, knowledge work and reasoning tasks, according to a blog post by Moonshot AI


Kimi K3 is designed for long-horizon coding, knowledge work and reasoning tasks, according to a blog post by Moonshot AI.


Moonshot AI unveils world’s largest open-source AI model as China narrows gap with US rivals

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Kimi K3 is designed for long-horizon coding, knowledge work and reasoning tasks, according to a blog post by Moonshot AI

Kimi K3 is designed for long-horizon coding, knowledge work and reasoning tasks, according to a blog post by Moonshot AI