Moonshot AI Releases Kimi K3, a 2.8 Trillion-Parameter Model

TL;DR. Moonshot AI has released Kimi K3, an open-weight multimodal AI model with 2.8 trillion parameters and a 1-million-token context window. - Kimi K3 uses a Mixture-of-Experts architecture with 896 experts, setting a new scale for open-source LLMs. - The model supports native text, image, and video processing, performing competitively against proprietary frontier models. - Developers can access Kimi K3 via multiple API providers or consider self-hosting with significant hardware investment.

• Moonshot AI released Kimi K3, a 2.8 trillion-parameter open-weight model.
• Kimi K3 features a 1,048,576-token context window and native multimodal capabilities (text, image, video).
• It employs a Mixture-of-Experts (MoE) architecture with 896 experts, leveraging Kimi Delta Attention and Stable LatentMoE.
• Kimi K3 benchmarks competitively against advanced proprietary models like Claude Fable 5 and GPT-5.6 Sol.
• Access is available through Hugging Face, OpenRouter, Fireworks AI, Baseten, Together AI, and Moonshot's own platform.

Back to QLANKR News