Moonshot AI says new Kimi model narrows China’s gap with US rivals

Chinese artificial intelligence startup Moonshot AI said Friday it has released Kimi K3, a model it says narrows the performance gap with top US systems and outperforms several rival models on selected tests. The launch moved Chinese AI equities, with Z.ai down 28%, MiniMax Group falling 16% and Alibaba shares losing 4% on Friday.

Moonshot said Kimi K3 remains behind Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol on overall performance. The Beijing-based company said, however, that the model beat Anthropic’s Claude Opus 4.8 and OpenAI’s GPT 5.5 on benchmarks covering areas including coding and general agents.

The company described Kimi K3 as China’s largest AI model to date, with 2.8 trillion parameters. In AI systems, parameters are the internal variables adjusted during training, and their number is one measure of the scale of a neural network, though benchmark results and deployment costs also affect commercial usefulness.

Benchmark claims and compute limits

Bank of America analysts led by Alex Liu said the release showed Chinese developers were still gaining capability despite constraints on access to hardware and computing capacity. “Despite persistent hardware/compute capacity constraints in China, K3 demonstrates that pre-training scaling, paired with architectural innovation, can still deliver step-change gains for flagship Chinese models,” the analysts wrote in a note.

Moonshot, founded in 2023 and based in Beijing, is among China’s more prominent AI model developers. Its release comes as US and Chinese technology companies compete to build advanced models, while governments weigh the strategic and commercial implications of the technology.

CNBC has reported that Chinese AI models have been gaining use among Western companies as their performance improves and their usage costs remain below those of the most advanced US lab offerings. CNBC has also reported that US lawmakers are examining ways to limit adoption of Chinese AI models by domestic companies.

Pressure on Chinese AI peers

The market reaction was sharp among other Chinese model developers. Z.ai, which released a new model in June, fell 28% on Friday. MiniMax Group, another Chinese AI model company, declined 16%.

“K3 raises the capability ceiling for China AI models, shifting the burden of proof to other independent AI labs,” Liu wrote.

Alibaba, which develops the Qwen family of models, also came under pressure. Its shares fell 4% Friday after having gained earlier in the week on news of a China partnership with Apple, according to CNBC.

Liu said Alibaba could still benefit from broader growth in AI training and usage through its cloud business, particularly given tight computing resources. He added that “Alibaba Qwen’s ‘open-source leader’ narrative may face some tests.”

The release gives investors another data point in assessing how quickly Chinese AI labs can close performance gaps while operating under tighter access to advanced chips. Moonshot’s own claims still place Kimi K3 behind the top systems from Anthropic and OpenAI overall, even as the company says it has surpassed some near-frontier US models in specific benchmark categories.

This story draws on original reporting from CNBC.