← sources

Qwen 3.8 Flash-Next compared to GLM 5.3 Flash

source: reddit (reddit:LocalLLaMA)published 2026-10-14 22:29ingested 2026-10-15 02:42

retained text

Anecdotally, I've noticed that QFN seems more collaborative than GLM. QFN seems willing to engage in conversation. It feels more like an assistant. GLM seems more action-oriented and less willing to converse. That is, I ask GLM a question to engage in discussion: GLM runs off searching the repo and then often re-engages in development without discussion.

I say this using a self-hosted QFN q4\_k\_xl versus GLM 5.3 Flash hosted on OpenRouter! 

I wonder if others have had similar experience? If so, have you found any particular prompting to cause GLM 5.3 Flash to me more interactive and collaborative? 

 I suppose I could try lifting some prompting out of pi-goal-x or similar as that seems to *mostly* succeed at forcing GLM 5.3 Flash to the table. However, I found that GLM will often take the information given as part of the process, run off, and implement most of the solution before even negotiating the goals!

document fetched 2026-10-15 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1x6533a/qwen_38_flashnext_compared_to_glm_53_flash