Moonshot AI releases Kimi K3 open-weight model

Moonshot AI has released Kimi K3, a massive 2.8 trillion-parameter artificial intelligence model featuring open weights. The model utilizes a Sparse Mixture-of-Experts (Sparse MoE) architecture, activating 104 billion parameters per token, and employs Kimi Delta Attention (KDA) to reduce the cost of long-context processing. It features a 1-million-token context window and can process text, images, and video.

Despite its technical capabilities, the release has faced significant criticism regarding accessibility and usability. While the open-weight nature distinguishes it from closed models like those from OpenAI, the model's scale makes it impractical for anyone without a massive data center. Additionally, users have reported frustration with aggressive usage quotas and high costs, with some claiming that even high-tier monthly subscriptions can be exhausted in a very short period.