Moonshot AI retires Kimi K2.5 model to focus on Kimi K3

Moonshot AI has officially retired its Kimi K2.5 model and the older moonshot-v1 series. Kimi K2.5 was a trillion-parameter Mixture-of-Experts (MoE) model trained on 15 trillion multimodal tokens, noted for its performance in coding, visual understanding, and agent tasks.

The company is shifting focus toward its more advanced models, including the Kimi K3. Released in July, Kimi K3 is a 2.8 trillion parameter open-source model, making it one of the largest open-source models globally. It utilizes advanced technologies such as MoonViT-V2 and Hybrid KDA-MLA, featuring a 1M context window. Developers are encouraged to migrate to Kimi K3, Kimi K2.6, or specialized coding models like Kimi K2.7 Code.