main
official config + model card · 64 native INT4 shards
MOONSHOT AI / AGENTIC MULTIMODAL / 2026

Kimi-K2.7-Code

TOTAL1.1Tparameters
ACTIVE32Bper token
DEPTH61 + ViT27repeated blocks
HIDDEN7,168main trunk
CONTEXT256Knative tokens
CHECKPOINT595.2GB64 shards
INTERACTIVE SYSTEM BLUEPRINT

MODEL MRI

Kimi K2.7 Code · Multimodal Agentic Explorer

100K INPUT + 1K OUTPUT · 19.21P arithmetic FLOPs · click any module for evidence

VERIFIED ADAPTER
INPUT100.0KTEXT + MEDIA
TRANSFORMER STACK61 TEXT LAYERS1 dense · 60 sparse MoE · MLA throughout
pre-norm additive residual · standard two-sublayer skip
++
OUTPUT1.0KTOKENS
wraps every repeated block
MULTIMODAL + MLA SYSTEMvision tokens become ordinary prefix tokens after projection
KIMI MoE · 1T TEXT BACKBONE32B active path · total capacity ≠ active compute
+
DEPTH61 + 27LLM + ViT
HIDDEN7,168text trunk
ATTNMLAdense read
EXPERTS384routed
ACTIVE8 + 1per token
PRECISIONINT4BF16 critical