Systems engineer Yacine shows GLM-5.2 outperforming GPT-5.5 at decompiling and simplifying assembly code
The model isolated and resolved complex integer-to-float bit-punning.
Users are excited about GLM-5.2 because it effectively handles complex assembly decompilation, reverse engineering, and bit-punning detection in legacy code.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@banteg Isn't it much slower than 5.5?
@mteamisloading true, it's slower, just showing it's not useless. seems like a decent model.
not a bad result from glm-5.2, it found this after gpt-5.5 xhigh
@yacineMTB wait this is actually insane
have you tried it yourself? curious how well it handles more complex stuff
@banteg gpt5.5 is very good at complex logic, not good at implementing with clean code principles. luckily, once implemented, a swarm of agents are good at cleaning this up.
@banteg did u run them in parallel for same thing ?
@banteg Is that Pi?
@yacineMTB yeah
@banteg gotta love when models finally get it right
@yacineMTB Emulators in shambles. Elon was right on the AI + assembly combo
@banteg @mteamisloading how much slower are we talking about vs codex (not on /fast)
@banteg Anf you can run it locally if you have the juice.
This is good
@yacineMTB Someone alert the FBI, this model could be used to decompile a person!!!
@yacineMTB im using it to write new assembly :) simd avx512 :)
@yacineMTB I've been using it for security red teams and Kimi for scraping because they're much less judgemental about these things
@yacineMTB 🤔 sometimes it refuses like opus 🤣 minimax m3 never does
@DarayuthH @yacineMTB I did it with the prior version …
@yacineMTB It’s a really powerful model. It’s easy to spin up by flock
https://github.com/hadihonarvar/flock
@yacineMTB are these results good, you being into robotics?