Beijing-based startup Moonshot AI has unveiled Kimi K3, an open-weight artificial intelligence model that the company says can compete with leading frontier AI systems developed in Silicon Valley.
The launch highlights the growing influence of Chinese AI companies in a market that has largely been led by US developers, while also increasing interest in open-weight models that can be customised and deployed locally.
Kimi K3 introduces a 3-trillion-scale architecture
Kimi K3 is the world's first open-weight AI model in the 3-trillion-parameter class, featuring a total of 2.8 trillion parameters. The model has been designed to support complex reasoning, multi-stage coding workflows, problem-solving, and other advanced cognitive tasks.
According to Moonshot AI, the model can scale efficiently while maintaining strong performance across coding and deep-reasoning applications with minimal human intervention.
Kimi K3 also features a context window of up to one million tokens, enabling it to process and analyse large volumes of information over extended tasks. This capability supports long-horizon coding pro-jects and continuous analytical workflows.
The architecture combines updated Kimi Delta Attention (KDA) and Attention Residuals (AttnRes) models to improve performance across longer sequences and deeper layers. Using the Stable Latent-MoE framework, Kimi K3 routes requests to 16 out of 896 available experts, helping optimise efficiency and resource utilisation.
Open-weight design offers flexibility for developers and enterprises
As an open-weight model, Kimi K3 can be downloaded, customised and deployed by organisations and developers on their own infrastructure.
This approach may appeal to businesses seeking greater control over their AI environments, particularly where data privacy, regulatory requirements and application customisation are priorities. Running AI workloads on local infrastructure can also help organisations optimise operational costs and tailor solutions to specific business requirements.
Advanced coding and agentic AI capabilities
Moonshot AI says Kimi K3 has been built to support multi-stage software engineering tasks, enabling it to navigate large codebases, coordinate terminal operations and maintain extended development workflows with limited supervision.
The company stated that the model successfully completed end-to-end production workflows during testing, demonstrating advances in autonomous reasoning and agentic AI capabilities.
What Kimi K3 could mean for the AI market
Kimi K3 is scheduled for full release on 27th July and has already attracted attention across the AI industry.
Its launch comes as discussions continue around AI governance, cybersecurity, and emerging regulatory frameworks worldwide. Because Kimi K3 is being released as an open-weight model, it enters a rapidly growing segment of the AI market that offers organisations greater deployment flexibility than many closed systems.
Benchmarking organisations Artificial Analysis and Arena.ai have reported that Kimi K3 performs competitively against leading models from OpenAI and Anthropic. In blind user-preference evaluations, the model reportedly outperformed Anthropic's Fable.
The longer-term impact of Kimi K3 will ultimately depend on enterprise adoption, developer uptake and how effectively it competes against established AI platforms. However, its arrival signals that competition in the frontier AI market is continuing to broaden beyond traditional industry leaders.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
Destination AI
Destination AI from TD SYNNEX is our comprehensive end-to-end programme designed to help you and your customers get ahead of the AI curve.
