Entropix creator xjdr praises GLM 5.2, while engineer Ben (no treats) prefers it over K2.7
Ben compared the models using the Devin software engineering assistant.
Many users praise GLM 5.2 for performing well in their workflows and evaluations while preferring it to alternatives like Kimi or Fable, though one user expressed frustration with it sitting idle on the cloud.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@_xjdr tried it and k2.7 thru devin and i think i liked glm more
y'all, glm 5.2 is very good
@JoshPurtell ya, im running it on my eval stack now but also i am using it instead of kimi as a test. so far, its performing incredibly well on both
@_xjdr it's so good i literally had to double check
@_xjdr Evals?
@_xjdr been saying
@_xjdr and it doesn't freak out and demote you for trying to talk about dragonyflies *glares at Fable*
@_xjdr sickos yes
@_xjdr Kanging about distilling aside doesn't it feel like a Claude?
CoT has stuff I don't normally see in opus outputs so definitely different.
@_xjdr i got myself specing out threadrippers for 8xB70s nodes to spin that up locally (just dont be poor i guess)
@_xjdr As good a GPT 5.4? Better?
@_xjdr It really is.
@_xjdr the fastest eval signal now is whose workflow a model survives in, not its leaderboard score. a careful builder saying 'it's very good' after actually using it carries more information than another benchmark number. adoption by skeptics is the eval that ships.
@_xjdr It sits idle upon cloud as I write this, giving me malcontent.