Chinese AI startup Moonshot will introduce the Kimi K3 language model in the coming days, which, according to a number of industry tests, can outperform the Opus 4.8 model from American Anthropic.
Chinese AI startup Moonshot will introduce the Kimi K3 language model in the coming days, which, according to a number of industry tests, can outperform the Opus 4.8 model from American Anthropic.
Kimi K3 will be the largest language model ever created in China, with 2 to 3 trillion parameters. For comparison, experts estimate the size of Opus 4.8 at 1.5 to 2 trillion parameters, although Anthropic itself does not officially disclose this data.
According to the FT, the Kimi K3 will outperform Opus 4.8 in key benchmarks and will be available for free download in open-source format, meaning users can modify the model to suit their needs. At the same time, the new Moonshot is likely to be inferior to Anthropic’s Fable model, which the company temporarily withdrew amid concerns from US authorities about its potential for cyberattacks.
According to the publication, the launch of Kimi K3 could shake the widespread belief that Chinese AI models lag behind American ones by 8-12 months. Open access to the model could also increase competitive pressure on Anthropic and OpenAI, which keep their most powerful models private.
The FT also notes that more and more companies are switching to cheaper Chinese models due to the rising cost of using American AI services — for example, the previous Moonshot model, K2.6, costs about three times less than Opus 4.8.
According to one of the publication’s interlocutors, Moonshot is currently raising a new round of funding at a valuation of about $31.5 billion. For comparison, DeepSeek is seeking a valuation of about $71 billion, while Anthropic and OpenAI are valued in the hundreds of billions of dollars.
Moonshot is one of China’s leading developers of generative artificial intelligence. In recent months, Chinese companies, including DeepSeek, Z.ai, and Moonshot, have been actively closing the technological gap with American competitors by offering open and cheaper models, which is intensifying competition in the global AI market.
Anthropic has previously accused some Chinese AI companies of distilling their models, a method that allows new models to be trained on the responses of more powerful systems. Some of the accused companies have rejected these claims, calling them an attempt to protect the monopoly position of American developers.
Comment hidden for violating commenting rules.