Moonshot releases Kimi K3 weights, a 2.8T MoE model with 1M-token context
Gradio · x · 2026-07-29
- Moonshot released the weights and technical report for Kimi K3.
- Kimi K3 is described as the company’s most capable model: a 2.8T MoE system with native visual understanding and a 1M-token context window.
- Moonshot also says the model uses a new architecture that delivers 2.5× more intelligence per unit of compute, and it is opening more of the stack around it, including high-performance attention kernels, an MoE communication library, and infrastructure for running agent environments at scale.
- The repost also points to an open-source Hugging Face workflow for trying Kimi K3 without API keys.
Related event: Moonshot AI Open-Sources Kimi K3: A 2.8T Parameter MoE with 1M Context(7 posts)→
More from coding & agent
- Codex community meetup set for July 29 in San Francisco with $100 credits for first 50 signups — paw_lean · 2026-07-29
- ComfyUI tutorial shows KREA 2 Identity Edit v1.2 for low-VRAM face editing — cgpixel23 · 2026-07-29
- Nanomi Work chained three models to cut a 32-second promo video in a real test — AlchainHust · 2026-07-29
- OpenCode meme mocks a 1,485-line feature diff as a “slop cop citation” — teropa · 2026-07-29
- OpenAI resets ChatGPT Work and Codex limits, extending typical usage by 18% — haider1 · 2026-07-29
- Opus 5 turns Spider-Man PS4 into a joke version called “spooderman” — repligate · 2026-07-29