Wow.
@Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose.
It responds with nuance and judgement, & handles long context VERY well.
I've never experienced an open weights model like this before.
@Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose.
It responds with nuance and judgement, & handles long context VERY well.
I've never experienced an open weights model like this before.
230
484
7,286
891,438
The one big gap is that it is blind - it can't handle images at all.
If they fix that, then perhaps this could become the best available model in the world.
If they fix that, then perhaps this could become the best available model in the world.
Jun 18, 2026 · 11:52 PM UTC
19
11
540
58,722
BTW I've been using using @FireworksAI_HQ for inference. Other providers might not be as fast.
15
13
377
50,788
@jeremyphoward You can add small vision model to work with, if running locally
2
0
4
573
@jeremyphoward I might be misquoting @mattpocockuk but it's no bad thing missing vision from an LLM like this. Use dictation and just describe what you need. Or something.
0
0
0
49
@jeremyphoward I mean, for that, it's easy to use a cheap vlm as a subagent for any requests that actually need image understanding, IMHO.
0
0
0
205
@jeremyphoward Ouch! What I am working on requires excellent image analysis as well as good reasoning.
0
0
1
198
@jeremyphoward ui aesthetic like this with no vision tower is so neat
0
0
0
309
@jeremyphoward you can use specialized model for image-to-word for context building.
0
0
1
112
@jeremyphoward one have to be really good at prompting to fullly milk it then
0
0
0
16
@jeremyphoward Something I used to do in the past is start a chat with a model with it's own mmproj, then switch models. The new model could answer questions about the image. As if by magic.
0
0
0
507
@jeremyphoward People are using tools to add image analysis capabilities to glm and it seems to work very well.
There is a nice discussion about it on redit reddit.com/r/opencode/s/D…
There is a nice discussion about it on redit reddit.com/r/opencode/s/D…
0
0
0
48
@jeremyphoward Maybe that's where agents could come in.
0
0
0
62
@jeremyphoward Oh well - you can still use gpt and opus for image to markdown — did you try Kimi 2.7 too?
0
0
0
467
@jeremyphoward Good image models are tiny and plentiful. Stop asking for multimodality and just allow a model to be good in its niche.
0
0
0
132
@jeremyphoward Good. strip it down to just coding/engineering/math/etc Make it run at home.
0
0
0
143
