Chinese artificial intelligence startup Moonshot has introduced Kimi K3, describing it as the world’s largest open-weight AI model.
The company said the model contains 2.8 trillion parameters, making it the first open-weight system to approach the three-trillion-parameter mark.
According to Moonshot, Kimi K3 is designed for advanced reasoning, long-horizon coding, and complex knowledge work while delivering performance close to that of some of the industry’s most advanced AI systems.
The launch reflects the rapid pace at which Chinese AI companies are expanding their capabilities as competition in the global AI industry intensifies.
What is Kimi K3?
Moonshot said Kimi K3 is its most capable flagship model to date.
The model is built on Kimi Delta Attention (KDA), a hybrid linear attention mechanism, along with Attention Residuals, which the company said improve computing efficiency.
The company added that Kimi K3 includes native visual understanding and supports a 1-million-token context window, allowing it to process and retain significantly more information in a single prompt than earlier generations.
According to Moonshot, the model is intended for frontier intelligence applications, including long-horizon coding, reasoning, and knowledge-intensive tasks.
The company also described Kimi K3 as the world’s first open-source model in the three-trillion-parameter class.
It represents the latest milestone in its effort to expand the scale of open-source AI models.
What are open-weight models?
Unlike proprietary closed-source AI systems, open-weight models allow users to download, run, and customise the underlying model.
This gives developers and businesses greater flexibility to adapt AI systems for their own applications instead of relying entirely on cloud-based proprietary services.
How does Kimi K3 compare with other AI models?
Kimi K3 delivered competitive performance against Anthropic’s Fable 5 model and substantially outperformed Anthropic’s Opus 4.8, GPT 5.6 Sol, and GPT 5.5 in GPU kernel optimisation.
The company said GPU kernel optimisation refers to techniques that maximise AI hardware utilisation while reducing latency.
Moonshot also highlighted results from several third-party evaluations.
Vals AI placed the model second overall behind Fable 5 and ahead of GPT-5.6 Sol.
Artificial Analysis also reported that Kimi K3 delivered performance comparable to OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8, particularly in tests involving complex, multi-step tasks.
What makes Kimi K3 different?
Moonshot said Kimi K3 introduces two major architectural upgrades aimed at improving computational efficiency and enabling long-horizon coding tasks with minimal human supervision.
The company also said Kimi K3 continues its effort to push the boundaries of open-source model scale.
According to Moonshot, Kimi models have remained at the frontier of open-source model scale during nine of the past twelve months.
What does the launch mean for China’s AI industry?
The unveiling of Kimi K3 comes as Chinese AI companies accelerate model releases and narrow the gap with leading US developers.
Companies including Moonshot, Z.ai, and MiniMax have been introducing increasingly powerful AI systems while reducing deployment costs, challenging the long-held view that Chinese AI developers significantly lag their US counterparts.
The latest launch follows the introduction of Z.ai’s GLM-5.2, which attracted industry attention after posting benchmark scores close to leading US closed-source AI models.
Investor reaction was mixed.
Following Moonshot’s announcement, shares of domestic AI rivals Zhipu and MiniMax fell sharply in Hong Kong.
Just before the market closed, the stocks were down 27.7% and 16.5%, respectively.
Moonshot, which is backed by Alibaba and Tencent, has continued expanding its AI capabilities and funding.
Bloomberg previously reported that the company was seeking $2 billion in fresh funding at a valuation of about $30 billion ahead of a potential Hong Kong listing.
The post Kimi K3 explained: Inside Moonshot’s biggest open AI bet yet appeared first on Invezz
