Moonshot’s Kimi K3 ships as a 2.8T open-weights MoE with 1M-token context
Latent Space · rss · 2026-07-28
Latent Space's AI News roundup argues that open weights dominated the week, with Moonshot's Kimi K3 as the standout release.
Kimi K3 ships as a 2.8T-parameter MoE with 104B active parameters, 896 experts, 16 active per token, a 1M-token context window, and native visual understanding. Moonshot also open-sourced FlashKDA, MoonEP, and AgentENV, making the release more than a model drop: it's a full stack for large-scale agentic post-training and serving.
The roundup says K3 reportedly improves scaling efficiency by about 2.5× over K2 and has been validated by multiple external evaluations as beating Opus 4.8 among open-weights models. It also notes that the license is source-available rather than permissive OSS, with commercial carve-outs for large hosts and high-revenue products. Distribution was broad on day one across vLLM, Baseten, Modal, Fireworks, Nebius, Together, DigitalOcean, Cursor, Cognition, Ollama Cloud, and Dell Enterprise Hub.
More from coding & agent
- Copperhead is an open-source AI agent for PCB design on local KiCad files — cneuralnetwork · 2026-07-28
- Micro Codex Mobile enters beta with chats, fast mode, and reasoning controls — Dimillian · 2026-07-28
- Most of my script failures were navigation-layer breakages, not logic bugs — Opening-Profile6279 · 2026-07-28
- A user says Opus 5 made 17 monthly exports unnecessary in a coding workflow — gaganghotra_ · 2026-07-28
- Most agent failures are really handoff failures to the human reviewer — Fit_Average8352 · 2026-07-28
- A design-to-engineering team built an eval to stop AI copy from shipping on vibes — Fit_Average8352 · 2026-07-28