China's Moonshot AI Unveils 2.8-Trillion Parameter Model to Challenge US Rivals

Chinese artificial intelligence startup Moonshot on Friday unveiled Kimi K3, a new open-source model it claims is the world's largest and performs close to leading U.S. frontier offerings. The release marks a significant step in China's effort to rival American dominance in the generative AI sector.

Model Specifications and Benchmarks

According to a company statement, Kimi K3 is a 2.8-trillion parameter model featuring a 1 million token context window and inherent multi-modal capabilities. Moonshot asserted that the model's performance is comparable to Anthropic’s Fable 5 and substantially outperforms OpenAI’s GPT 5.6 Sol and GPT 5.5 in certain benchmarks.

Separately, the benchmarking platform Arena.AI reported that Kimi K3 ranked first in a frontend coding test, placing it ahead of the previous leader, Claude Fable 5. These claims position the model as a serious contender in the high-stakes global AI race.

Competitive Landscape

The launch of Kimi K3 underscores a broader push among Chinese AI companies to develop competitive, and in some cases superior, alternatives to U.S. technology. The release follows recent high-performance models from other Chinese firms like Zhipu and DeepSeek, many of which have also been open-source.

Ad

This trend aligns with a wider strategic initiative by Beijing to promote local AI development and build a self-sufficient technology ecosystem.

Investor Backing and Valuation

Moonshot's rapid development is backed by some of China's largest tech companies, including Alibaba (NYSE:BABA) and Tencent (HK:0700). The startup's progress has attracted significant investor attention.

A Bloomberg report last month stated that Moonshot was seeking to raise $2 billion at a valuation of approximately $30 billion. The report also noted that a Hong Kong initial public offering was under consideration, highlighting the firm's substantial growth ambitions.