China’s Moonshot unveils AI model, fuelling tech rout
Sign up now: Get insights on Asia's fast-moving developments
Moonshot released Kimi K3 on July 17, touting benchmarks comparable with some of the US labs’ best offerings.
PHOTO: REUTERS
- Moonshot unveiled its AI model Kimi K3, claiming performance close to top US models and surpassing local rivals, signalling China's growing AI capabilities.
- Kimi K3 is open weight with 2.8 trillion parameters, priced competitively and outperforming most rivals except the very top US models.
- The launch caused a tech stock sell-off amid concerns over competition, while Beijing boosts AI development and local firms plan IPOs and global expansion.
AI generated
Chinese AI pioneer Moonshot unveiled a new model that it claims performs on a par with some of the top-tier platforms from OpenAI and Anthropic, the latest sign that the Asian country’s artificial intelligence labs are closing a technology gap with the US.
Moonshot released Kimi K3 on July 17, touting benchmarks comparable with some of the US labs’ best offerings.
The model is open weight, meaning its parameters are available for users to download and customise, and outperforms all rivals except Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 on overall capability, it said.
Moonshot’s latest platform underscores how rapidly China’s AI race is evolving beyond a competition on price alone. Moonshot priced Kimi K3 at roughly Anthropic Sonnet levels, showing the company believes it can charge a premium to other Chinese models given its capabilities.
The start-up also claimed it surpasses local rival Z.AI’s most advanced offering on coding tasks. Those are among the most lucrative offerings by advanced model companies, driving revenue for Anthropic and OpenAI ahead of their planned initial public offerings (IPOs).
If Kimi can undercut them on price and achieve similar results, that could upend the business case for its rivals in the US and China.
Shares of its Chinese rival tumbled 28 per cent in their biggest slide since listing, while MiniMax Group slid 16 per cent.
The declines came amid a broader sell-off in Asian technology stocks, with investors concerned about whether a years-long AI-fuelled stock rally has become overstretched. The tech-heavy Nasdaq 100 Index closed down 1.5 per cent, finishing its worst week in almost a month.
The K3 model is “brilliant”, said Leonid Mironov, a portfolio manager at Gavekal Capital. “In my use, it’s clearly the best Chinese model ever.”
The performance of Kimi K3 came as a surprise to the market and raised concerns over Z.AI’s competitive position.
Moonshot said the model has 2.8 trillion parameters and a one-million-token context window, gauges of its capability.
Artificial Analysis ranked Kimi K3 ahead of Anthropic’s Opus 4.8 on some frontier benchmarks, making it the first Chinese open-weight model to achieve that milestone.
Whenever there is a new AI model, the existing ones get hit because of concerns about competition, said Steven Leung, executive director at UOB Kay Hian.
The move also comes as Beijing steps up support for domestic AI development, with President Xi Jinping hailing China’s advances in low-cost artificial intelligence while calling for a more open global technological order.
Chinese models have already put pressure on the frontier models by offering pricing that is far below what is available from OpenAI and Anthropic.
The weighted average cost of performing a standardised intelligence task using DeepSeek’s V4 Flash model costs just two US cents (three Singapore cents), compared with US$2.75 for the same task on Anthropic’s Claude Fable 5, according to benchmarking site Artificial Analysis.
DeepSeek just completed a record US$7.4 billion fund raising and is planning an IPO in 2027, giving it more capital to develop AI services and offer them globally at prices far below US rivals.
Meanwhile, Z.AI is on track for annual recurring revenue of US$1 billion, people familiar with the matter said. BLOOMBERG
