Kimi K3 开源了，2.8T MoE 模型带视觉和超长上下文，还开放了底层工具链，值得研究。
Kimi AI 开源了其最强模型 Kimi K3 的权重和技术报告。K3 是一个 2.8T 参数的 MoE 模型，具备原生视觉理解和 1M token 上下文窗口。新架构使每单位计算智能提升 2.5 倍。此外，Kimi 还开源了高性能注意力内核、MoE 通信库和智能体环境基础设施。模型权重和报告已在 Hugging Face 和 GitHub 公开。
Open Frontier Intelligence! Kimi K3 open weights and technical report have arrived! Recommended re...
Open Frontier Intelligence! Kimi K3 open weights and technical report have arrived! Recommended reading. Kimi.ai @Kimi_Moonshot Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Kim… Tech report: github.com/MoonshotAI/Kim… Tech blog: kimi.com/blog/kimi-k3 🔗 View Quoted Tweet 💬 1 🔄 2 ❤️ 12 👀 2080 📊 4 ⚡