Verified75% confidenceFactExact time
Kimi K3 uses a MoE (Mixture of Experts) architecture, configured with 896 experts, with 16 activated each time.
3
Sources
75%
Confidence
Long-term
Relevance
7/22/2026
First Seen
Sources
Related Claims
VerifiedKimi K3 almost certainly adopts a Mixture of Experts (MoE) architecture.85% similarUnverifiedKimi K3.1 uses a MoE architecture, with about 2.8 trillion total parameters and 104 billion activated parameters.74% similarUnverifiedMoE is a sparse activation architecture; DeepSeek V3 has 256 experts but activates only 8 at a time.69% similarUnverifiedHY3 adopts a Mixture of Experts (MoE) architecture with a total of 295 billion parameters, 21 billion activated parameters, 192 experts, and a Top-8 routing mechanism.69% similarUnverifiedAccording to a leak by Bilibili creator Su Su, the Kimi official website's frontend model-selection code contains K3D1 Agent (i.e., the Kimi K3.1 agent version), and OK Computer additionally includes a Make 3D mode; the leak judges that K3.1 has entered gray release or is about to launch.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/590285API
curl https://kongchang.com/api/v1/knowledge/claims/590285MCP
get_claim(id=590285)