main
official Qwen3.6 config + 26 BF16 shard headers
QWEN / HYBRID MULTIMODAL / 2026

Qwen3.6-35B-A3B

TOTAL35.95Bparameters
ACTIVE3Bper token
DEPTH40repeated blocks
HIDDEN2,048main trunk
CONTEXT262Knative tokens
CHECKPOINT71.92GB26 shards · Apache-2.0
INTERACTIVE SYSTEM BLUEPRINT

MODEL MRI

Qwen3.6-35B-A3B · Architecture Explorer

100K INPUT + 1K OUTPUT · 1.45P arithmetic FLOPs · click any module for evidence

VERIFIED ADAPTER
INPUT100.0KIMAGE · VIDEO · TEXT
TRANSFORMER STACK40 HYBRID LAYERS30 DeltaNet · 10 full attention · MoE throughout
RMSNorm Residual
++
OUTPUT1.0KTOKENS
wraps every repeated block
GATED DELTANET · HYBRID SCHEDULE周期性全局 Attention 修补纯线性状态的直接检索能力。
Qwen MoE · 35.95B CAPACITY3B active · total capacity ≠ active compute
+
DEPTH4030 + 10
HIDDEN2,048main trunk
HEADS16 / 2Q / KV
EXPERTS256top-8
CONTEXT262Ktokens
PARAMS35.95Bheader