And it’s only 40B active / 744B total params…
Hunter Bown@goodhunt
still can't believe how good glm 5.2 is
12:48 PM · Jun 20, 2026 · 20.1K Views
Hugging Face co-founder Thomas Wolf highlights GLM-5.2's MoE architecture, prompting requests for a sparse SmolLM4
GLM-5.2 activates 40 billion of 744 billion parameters.
918741620.6K
Sentiment
Users praise Zai_org for consistently crushing it with GLM-5.2's strong results at efficient 40B active parameters.
Pos
100.0%
Neg
0.0%
1 comments with sentiment.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Posts from X
Most Activity
Most Activity
VIEWS642LIKES10
elie@eliebakouch
@Thom_Wolf smollm4 40B / 744B wen? 🥺
Thomas Wolf@Thom_Wolf
And it’s only 40B active / 744B total params…
5hViews 642Likes 10Bookmarks 0
REPLIES1
Samuel Ekpe@samuelekpe
@Thom_Wolf What are the inactive ones doing?
6hViews 104
Tyler Williams@unmodeledtyler
@Thom_Wolf @Zai_org is absolutely onto something. I’ve thought that since 4.5 air - they just keep crushing it.
7hViews 105Likes 2
wafer@wafer_ai
@Thom_Wolf we made glm 5.2 pretty fast btw
7hViews 147Likes 1
Hunter Gon@gonlenidefi
@Thom_Wolf being surprised at those numbers feels like we havent hit ceiling yet
somewhere a 100B just dropped and nobody blinked
8hViews 107Likes 1
Solomon@solomonbotchway
@samuelekpe @Thom_Wolf
6hViews 18Likes 2
Da7em@Da7_Tech
@Thom_Wolf
7hViews 128
tsunami_crypto@ls_brd
@Thom_Wolf smaller active params usually means more distilled optimization
kinda wild if it actually holds up
8hViews 126