Chinese company Moonshot AI has announced the scheduled release of its new flagship model, Kimi K3, on July 27, 2026. The model represents the world's first open frontier 3T-class (3 trillion parameters) system, designed to solve complex coding tasks and deep data analysis.
What Happened
On July 27, 2026, Moonshot AI plans to publish the weights for the Kimi K3 model. This 3-trillion-parameter model uses the innovative Kimi Delta Attention and Attention Residuals architecture, allowing it to work efficiently with an expanded context window, including the analysis of entire code repositories.
Context
The release of 3T-parameter scale models as open weights marks a transition toward the "open frontier" category. This places open-source developments on par with closed proprietary systems like GPT-4 and Claude 3.5, significantly changing the landscape of available technologies.
Why It Matters for the Industry
For the industry, this signifies a new technological standard in the open-source segment. The use of Delta Attention architecture may stimulate the development of new efficient attention methods for ultra-large models and increase demand for specialized hardware for 3T+ scale model inference.
Why It Matters for Users
Developers and researchers will gain access to a world-class tool for automating code writing and working with massive volumes of knowledge. The ability to deploy locally on private servers or powerful GPU clusters will ensure data security and control while using models at the level of top-tier closed APIs.
What Is Currently Unknown / Limitations
At this time, detailed technical specifications regarding inference requirements are missing, which causes skepticism regarding the practical feasibility of deploying the model in production environments without significant computational resources.
Sources
Author
Look at AI, Editorial Team