Chinese AI startup Moonshot has unveiled Kimi K3, a 2.8 trillion-parameter open-weight model that it says is the largest of its kind and one capable of competing with the world’s leading frontier AI systems. 

The release marks another milestone in China’s rapidly advancing open AI ecosystem and adds to mounting evidence that the country’s developers are closing what was once considered a substantial performance gap with their US counterparts.

The announcement comes only weeks after the US government withdrew Anthropic’s Fable and Mythos models over security concerns, creating an unusual moment in the frontier AI landscape. While US companies continue to dominate the closed-model market, Chinese developers are increasingly differentiating themselves by releasing highly capable open-weight systems that can be downloaded, modified, and deployed by enterprises.

Moonshot describes Kimi K3 as the first open-weight model to approach the three-trillion-parameter threshold. Built for advanced reasoning, long-horizon coding, and knowledge-intensive work, the model also supports a one-million-token context window, allowing it to process and retain significantly larger volumes of information than previous generations within a single prompt.

The company’s claims extend beyond scale. Moonshot said Kimi K3 delivers competitive performance against Anthropic’s Fable 5 while outperforming Anthropic’s Claude Opus 4.8, OpenAI’s GPT-5.5, and GPT-5.6 Sol on GPU kernel optimization, a measure of how efficiently AI software utilizes computing hardware to minimize latency and maximize throughput.

Independent benchmark providers have also ranked the model among the industry’s top performers. Arena.ai ranked Kimi K3 first for web interface-building capabilities, while Vals AI ranked it second overall, behind Fable 5 but ahead of GPT-5.6 Sol. Artificial Analysis reported performance comparable to OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8, particularly on complex, multi-step reasoning tasks.

The launch reinforces a critical change in China’s AI industry. Companies including Z.ai, MiniMax, DeepSeek, and Meituan have accelerated their release cycles over the past year, steadily raising model capabilities while lowering deployment costs. That strategy is beginning to challenge a long-standing assumption among Western analysts that Chinese AI systems lag behind leading US models by several months.

Earlier this year, Z.ai’s GLM-5.2 surprised researchers by achieving benchmark results close to top American closed-source models, prompting renewed debate over how quickly China’s AI ecosystem is catching up.

Cost remains one of China’s strongest competitive advantages. Lian Jye Su, chief analyst at Omdia, told Reuters that Chinese models are attracting enterprise interest because they can be deployed far more cheaply than comparable US offerings. “They can be run at a fraction of the cost that OpenAI charges its clients,” he says, while cautioning that sheer parameter count should not be interpreted as a direct indicator of model quality.

That caveat may prove particularly relevant for Kimi K3. Despite being open-weight, its enormous size makes self-hosting impractical for most organizations. Ryan Fedasiuk, a fellow at the American Enterprise Institute, estimated that running a model with 2.8 trillion parameters locally would require computing infrastructure costing hundreds of thousands of dollars.

Moonshot said Kimi K3 introduces architectural improvements designed to increase computational efficiency and enable long-horizon coding tasks with minimal human intervention. The startup, backed by Alibaba and Tencent, has been expanding aggressively amid intensifying competition.