spiffytech Recently GLM 5.3 Flash has gone from usable to now it's so slow I just give up and walk away for minutes waiting for it to answer simple questions. Model responds quickly enough that I feel comfortable waiting on it.
shurik same here - it is painfully slow. takes at least 3 (!) minutes to answer a simple question when using the kagi assistant app on ios with GLM 5.3 Flash.
comicfrieze It's not just GLM 5.3, but I've confirmed using QWEN 3.8 27b and DeepSeek V4.1 Flash. It's painfully slow, if it doesn't time out completely. I've noticed this over the past 2 days or so (definitely September 16... maybe September 15).
mb comicfrieze With Qwen3.8 27B what takes time is reading the sources. If you disable web access it'll reply faster than you can blink, since it's hosted by Cerebras on their 27 kW mega processors.
