Moonshot releases open-weight Kimi K3 with 2.8T parameters and 1M context
ns123abc · x · 2026-07-27
Moonshot AI has released Kimi K3 as an open-weight, native multimodal agentic model.
- Scale: 2.8T parameters, with a 1-million-token context window.
- Architecture: built on Kimi Delta Attention (KDA), Attention Residuals (AttnRes), and a Stable LatentMoE setup that activates 16 of 896 experts.
- Capabilities: positioned for long-horizon coding, knowledge work, and reasoning, with text, image, and video support in one model.
- Release: the full weights are available under the Kimi K3 License for research, deployment, and further development.
Related event: Moonshot Open-Sources Kimi K3: A 2.8T Parameter Multimodal Model(19 posts)→
More from Models
- Kimi K3 launches on SGLang with 423 tok/s and 11 cloud partners — ying11231 · 2026-07-28
- Ollama Adds Kimi K3: 1M Context Window and Native Vision Support — ollama · 2026-07-27
- Kimi K3 reportedly improves training efficiency by 2.5× — zephyr_z9 · 2026-07-27
- NVIDIA distills Cosmos3 Super image-to-video to 4 steps with a 64B model — multimodalart · 2026-07-27
- Kimi K3 goes live on Modal with custom DFlash speculative decoding for lossless speedup — AAAzzam · 2026-07-27
- Kimi K3 with 2.8T parameters and 1M context now supported on vLLM — ricklamers · 2026-07-27