Kimi K3 Is Now Open-Weight: The Full Breakdown of the 2.8T Frontier Model
Summary
Moonshot AI released the full weights of its Kimi K3 model on HuggingFace on July 27, 2026, marking it as the largest open-weight model to date with 2.8 trillion parameters. This native multimodal model, developed by the Beijing-based AI lab, features a 1-million-token context window and introduces a new Kimi Delta Attention architecture. Kimi K3 utilizes a Mixture-of-Experts (MoE) design with 896 experts, activating 16 per token for an effective 104 billion parameters per forward pass. Moonshot AI describes Kimi K3 as their "most capable model to date" and the "world's first open 3T-class model," fulfilling a promise made on July 16, 2026.
Key takeaway
For AI Engineers evaluating large open-weight models for deployment, Kimi K3's 2.8 trillion parameters and 1-million-token context window offer unprecedented scale and capability. You should investigate its native vision support and MoE architecture for developing advanced multimodal applications, especially if your projects require extensive context or integrated image processing. Consider its availability on HuggingFace for immediate experimentation and integration into your workflows.
Key insights
Moonshot AI's Kimi K3 is the largest open-weight, native multimodal model, setting a new benchmark for frontier AI.
Principles
- Open-weight models are rapidly advancing in scale.
- MoE architectures enable massive parameter counts.
- Native multimodal support is a key capability.
In practice
- Explore 1-million-token context for long-form tasks.
- Utilize native vision for integrated multimodal applications.
Topics
- Kimi K3
- Open-weight Models
- Moonshot AI
- Multimodal AI
- Mixture-of-Experts
- Large Language Models
- HuggingFace
Best for: AI Architect, NLP Engineer, Computer Vision Engineer, AI Scientist, Machine Learning Engineer, AI Engineer
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by LLM on Medium.