K3Nova
Sparse routing
Each token is routed through selected experts rather than every expert, which changes compute cost per token.
Kimi K3 MoE terms
| Term | Kimi K3 value | Plain meaning |
|---|---|---|
| Total parameters | 2.8T | Full model scale |
| Activated parameters | 104B | Approximate active portion per token |
| Experts | 896 | Specialized parameter groups |
| Selected experts per token | 16 | Experts routed for each token |
| Context length | 1,048,576 tokens | Maximum listed context window |
Plain answer
Kimi K3 uses a Mixture-of-Experts architecture. Official materials list 896 experts, 16 selected experts per token, 2 shared experts, and 104B activated parameters out of 2.8T total parameters.
MoE improves efficiency because only part of the model is active for a token, but it does not make the full model small. The system still needs access to the full expert set.
Decision checklist
Use this section to keep the next step practical before you touch accounts, files, infrastructure, or hardware.
| Check | Look for | Why it matters |
|---|---|---|
| Read the official source | Start with the model card, help center, docs, or official product page before trusting summaries. | Model names, limits, and prices change. |
| Name the path | Decide whether you need a hosted app, API call, local client, or full self-managed deployment. | Different paths have different costs and account boundaries. |
| Check the boundary | Keep login, payment, API keys, private files, and account settings on the official surface. | A public guide should help you choose, not handle sensitive actions. |
| Keep evidence | Save the source link, date, version, and practical next step that shaped your decision. | This makes later review easier for a team. |
How to read the facts
Shared experts
Shared experts provide common capacity alongside routed experts.
Hardware implication
The full expert set still has to be stored, so deployment planning remains memory-heavy.
Quick scan cards
Total parameters
2.8T Full model scale
Activated parameters
104B Approximate active portion per token
Experts
896 Specialized parameter groups
Selected experts per token
16 Experts routed for each token
Context length
1,048,576 tokens Maximum listed context window
FAQ
What does MoE mean for Kimi K3?
It means Kimi K3 routes each token through selected experts instead of activating the entire model.
Does MoE reduce storage needs?
No. It reduces active compute per token, but the full expert set still needs to be stored.
Why does MoE matter for self-hosting?
Routing across experts can make interconnect and serving software as important as raw VRAM.