Moonshot releases Kimi K3 on Hugging Face as a 2.8T open-weight multimodal agent
iamfakhrealam · x · 2026-07-27
Moonshot AI has released Kimi K3 on Hugging Face as an open-weight, native multimodal agentic model.
- The model is described as Moonshot’s most capable system so far, with 2.8T parameters built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes).
- It supports native vision and a 1-million-token context window.
- Moonshot says K3 is designed for long-horizon coding, knowledge work, and reasoning.
- The post also highlights a Stable LatentMoE setup that activates 16 of 896 experts, which Moonshot claims improves scaling efficiency by about 2.5× versus Kimi K2.
- The report emphasizes open frontier weights for research, deployment, and further development.
Related event: Moonshot Open-Sources Kimi K3: A 2.8T Parameter Multimodal Model(19 posts)→
More from coding & agent
- Opus 5 agents reportedly wrecked projects, so the prompt got much harsher — doodlestein · 2026-07-27
- NVIDIA releases Molt, a PyTorch-native framework for agentic RL — dair_ai · 2026-07-27
- iLands says its agents formed an economy during a three-week closed beta — testingcatalog · 2026-07-27
- AI One-Shots 25k Lines of Code to Simulate Black Hole Electrodynamics in 5 Minutes — kylekabasares · 2026-07-27
- ComfyUI adds Comfy MCP so agents can build workflows from prompts — PurzBeats · 2026-07-27
- Wiz says Atlas tops CyberGym with 90.9% and found 200+ new bugs — vijaybolina · 2026-07-27