Chinese startup Moonshot unveils world's largest open AI model, closing in on US competitors

Chinese AI startup Moonshot AI has announced Kimi K3, currently the world's largest open-source language model with 1 trillion parameters. This marks the first time a Chinese company has created an open-source model of such scale, allowing other developers to freely use and fine-tune it. Kimi K3 is built on a Mixture-of-Experts (MoE) architecture, where the active parameter count per query is approximately 270 billion, making the model both efficient and powerful. The model was trained on the Nexa AIDC supercomputer with a cluster of 10,000 accelerators, requiring significant investment. Moonshot aims to follow the example of successful American counterparts, such as Meta's Llama, which have proven that an open-source model can become an industry standard and achieve commercial success. Although Kimi K3 currently lags behind American competitors in raw metrics, its parameters are close to leading closed-source models, indicating a quickly narrowing gap. The company has also provided developer tools to run the model on various devices.