Kimi K3 open-sourced: 2.8T-param MoE with native vision and 1M context window

Kimi K3 开源：2.8T MoE 模型与技术报告

Kimi open-sourced K3, its strongest model: a 2.8T-param MoE with native vision and a 1M-token context window. The new architecture claims 2.5× intelligence per unit of compute. Weights, high-performance attention kernels, an MoE communication library, and a large-scale agent runtime are all released. The post doesn't disclose training data, benchmark scores, or the license.

Why it matters: Moonshot open-sourced K3 with full weights, high-perf attention kernels, MoE comms library, and an agent runtime — not just a model dump. 2.8T MoE, 1M context, native vision, and a 2.5x compute efficiency claim make this a strong signal. Not scoring 90+ because we only have th...