← All models

GLM · Last 7 days

GLM 5.3

0–100 · higher means more positive community experience, not benchmark performance.

Updated 2026-09-05 14:21 UTC · 2026-08-29 – 2026-09-05 UTC
Community score67.5/100136 comments in 7 days

Community reviews

Selected comments from the last 7 days. The balance of excerpts does not represent the share of positive reviews.

9 selected excerpts

NegativeSafety & refusals

“Compared to other AIs, after this update and the arrival of GLM-5.3, it has a very aggressive security protocol. Anything I ask her to do or create, she gives off a completely distorted view, saying that it's dangerous and could cause risk”

NegativeGeneral text

“For example, GLM-5.3's real generalization ability still has a gap compared to overseas top-tier models, but its AA benchmark total score closely trails GPT-6 (only one point behind).”

ZhihuzhMachine translated
Show original

比如 GLM-5.3 的真实泛化能力与海外顶尖梯队仍有差距,但 AA 榜总分却紧咬 GPT-6(只差一分)

NegativeReasoning

“it had no problem completely busting a comprehensive GLM 5.3 diagnosis for me 5 minutes ago to which GLM had to admit defeat lol”

NegativeLocal deploy

“With GLM 5.3, your setup would do worse than what you see in benchmarks, since you'd have to run a lower quant than what the benchmarks use.”

NegativeLocal deploy

“You can only fit GLM 5.3 in Q4, so I doubt it makes sense to trade that for full precision on DeepSeek.”

NegativeReasoning

“Knowledge-rich: I had it transcribe a piece of music for Jasmine Flower, and the result was about 80-90% accurate. Give the same task to GLM-5.3, and it only repeats the first line before forgetting the rest.”

ZhihuzhMachine translated
Show original

知识丰富:我让它默写了一首茉莉花的音乐,做出来八九不离十。同样的任务交给 GLM-5.3,只会重复第一句,后面都想不起来了。

NegativeCoding

“geometry and polygons were weak”

NegativeGeneral text

“Both have objectively terrible staircases.”

NegativeCoding

“5.3 Flash is awesome, actually feels like the next generation compared to the bigger 5.3.”