Beijing-based Moonshot AI has released Kimi K3, an open-weight artificial intelligence model featuring between 2.8 trillion [1] and 2.88 trillion [2] parameters.
The release marks a significant escalation in the global AI race, as China seeks to produce high-performance systems that can compete with leading U.S. models despite restrictive export controls on computing hardware.
Kimi K3 is described as the largest open-source model ever created [3]. According to technical reports, the system achieved a 2.5-fold improvement in scaling efficiency compared to its predecessor, K2 [2]. This efficiency allows the model to operate at a massive scale while managing the compute-capacity limits currently facing Chinese developers [4].
In performance testing, the model has shown strengths in technical tasks. Kimi K3 outperformed Anthropic’s Claude Fable 5 on the Frontend Code Arena benchmark [1, 5]. Other tests indicate that the system's performance is comparable to or exceeds that of GPT 5.6 Sol [6]. While some reports suggest the model beats Fable 5 [5], other data indicates it comes close to that system across several different benchmarks [7].
Moonshot AI designed the system to provide the first open 3-trillion-class system to the public [1]. By releasing the model with open weights, the company allows developers to examine and build upon the architecture, a move that contrasts with the closed-source approach of many U.S. frontier labs.
The development of Kimi K3 occurs as Beijing continues to navigate U.S. export-control measures intended to limit China's access to the most advanced semiconductors. By optimizing scaling efficiency [2], Moonshot AI aims to maintain parity with Western AI capabilities using available hardware resources [4].
“Kimi K3 is described as the largest open-source model ever created.”
The arrival of Kimi K3 suggests that Chinese AI firms are finding ways to bypass hardware shortages through architectural efficiency and massive parameter scaling. By utilizing an open-weight strategy, Moonshot AI is not only challenging the market dominance of U.S. firms like OpenAI and Anthropic but is also attempting to establish China as the primary hub for the open-source AI community.
