Kimi has launched Kimi K3, a 2.8 trillion-parameter AI model with native vision capabilities and a long context window. This model aims to improve performance in coding and reasoning tasks, marking a significant advancement in open-source AI capabilities.
Kimi has introduced Kimi K3, claimed to be their most capable model yet. With its 2.8 trillion parameters, it is the first open 3T-class model designed for advanced intelligence tasks like coding, knowledge work, and reasoning. Immediate availability of Kimi K3 is on Kimi's platforms including Kimi.com, Kimi Work, and Kimi API.
Kimi K3 is constructed using new architectural elements such as Kimi Delta Attention and Attention Residuals, aimed at improving data flow across model operations. This model also features significant advancements in Mixture of Experts (MoE) technology, activating only a fraction of experts during tasks to enhance efficiency. Overall, Kimi K3 is approximately 2.5 times more efficient than its predecessor, Kimi K2.
Though Kimi K3’s performance is noted to be below proprietary models like Claude Fable 5 and GPT 5.6 Sol, it achieves leading results in its evaluation suite, outperforming numerous other models. This positions it competitively within the open-source AI landscape.
The model launches with a maximum thinking effort standard, and forthcoming updates will introduce various operational modes. Kimi is collaborating with partners to ensure robust deployment and aims to release full model weights by July 27, 2026. Further technical insights will accompany a detailed technical report on Kimi K3.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
Kimi has launched Kimi K3, a 2.8 trillion-parameter AI model with native vision capabilities and a long context window. This model aims to improve performance in coding and reasoning tasks, marking a significant advancement in open-source AI capabilities.