GLM-5 Reasoning

zhipu DENSE text MIT 2026-03-15

Architecture

Total params
32 B
Active params
32 B
Layers
60
Context
128 k

Detailed specs

Hidden size
5120
FFN size
17920
Attention heads
40
KV heads
8
Head dim
128
Vocab size
151552
Attention type
mha

Operator decomposition (per token)

Operator FLOPs / token Bytes / token
matmul 3.30e+10 3.30e+10
attention 8.81e+9 1.36e+10
rmsnorm 3.07e+6 1.23e+6

Compatible hardware