Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- "GLM-5.2 hit a problem here, because it can't read images. It isn't multimodal. So instead of looking at a screenshot, it fell back on a hacky workaround: it wrote scripts to read the raw pixel data and check whether the colors came out roughly as expected."
A better way would be to use https://github.com/openbmb/MiniCPM-V
by js4ever - Right, just give the text llm access to a vision specific agent and that problem can be solved. Or if you really want let it even call Opus with an image - seems like you’d still save money
- I’m actually amazed at the output since GLM doesn’t have eyes. If GLM 5.2 costs 1/5 as much, seems like it could be set up to reach out to a multimodal model for vision tasks when required. Closer to parity but probably still significantly cheaper.by toddmorey
- Yes, it could just make one call to a multimodal llm to describe the scene
- I'm also very impressed at the output given the lack of image support.
They picked a task that heavily favors a model that can do multi-modal with images, and GLM still came within striking distance.
What I'm hearing from this article is that the next generation of open models that includes better multi-modal support are basically no-brainers for adoption.
Seems like a HUGE win for Z.ai and open models in general here.
by horsawlarway - So GLM emits fewer tokens and does fewer tool calls, but still takes over twice as long to complete.
Can someone explain to me where that time usage is coming from if not from the model operation itself?
Are the individual tool calls more complex and take more time to complete? Or is the rate of tok/s lower because the model does more compute per token?
by xg15 - Could just be infra. I'm betting Anthropic is much better prepared.
- Probably the data center where the model is running more than anything. Another option is if Opus is using anything like a Mixture of Experts approach, in which case the amount of the model loaded in memory at one time could be smaller than GLM.
- I have noticed that Opus and GPT 5.5 are very good at adjusting their thinking / reasoning intensity depending on the task at hand, something the open weights models are still not as good at.
In addition to that, some of the open weights models like GLM 5.2 or DeepSeek v4 Pro tend to be MUCH slower when generating tokens, which contributes to the perceived slowness. Although I wouldn't call models like GLM 5.2 slow by any means, e.g. it is currently one of the fastest models inside Notion today.
by iagooar - > Opus 4.8 built in Claude Code; GLM-5.2 built in Pi over OpenRouter.
It would be more interesting and accurate to see the comparison on the same harness if the intent is to compare the frontier models.
Pi is relatively new and does not have many features built-in compared to Claude Code. It was chosen intentionally this way as Pi's goal is not to create a bloat builtin of tools most don't use but to allow the users to customize to fit their need -- similar to Neovim vs IDE.
The end-user "vibe coding" experience is *heavily* swayed by the harness because prompt effectively drives how a model outputs an answer.
by jameson - > Through an API it costs a fraction of Opus, and you can run it yourself for free if you have the hardware.
I haven't been keeping up on hardware costs for state of the art LLM inference, but this remark made me ask myself how many readers of the article would actually be able to run this model on hardware they own. How much would it cost to acquire such a setup?
- Practically nobody.by crimsoneer
- The price of a small house.by bestouff
- GLM-5.2 performing like it would from a good provider - 8x B200s, so $450k. (No personal experience here)
GLM-5.2, severely quantised, 512GB Mac Studio, somewhere between $10k-$35k for a used M3. Or run it on a CPU with 768GB of RAM by getting an old PowerEdge with DDR4 for around $5,000.
Qwen-3.6-35b-q6, runs well on an RTX 5090 ($4000 + cost of a PC), runs medicore on an Intel Arc B70 ($1000 + cost of a PC plus lots of fiddling to get the setup to work right).
Gemma is a good candidate for the cheaper stuff, but I lack personal experience with using it locally
by trollbridge - This framing local LLMs as free is stupid. Basically pay 100+ months worth of API costs up front isn't free in the slightest. And it will be slower than non-local, your hardware will be outdated in 12 months and probably won't be able to run SOTA at anywhere near non-local speed in max 20 monthsby jack_pp
- I've signed up with Ollama to experiment with these open source models. For the past 3 months, it's just been experimenting, trying it out. GLM is the first model that I am using on a daily basis to do my coding work (as well as using Claude). It's good - I've been maxing out my Ollama usage limits everyday :)by postatic
- Cool to hear, what kind of tasks have you been using GLM for? And what other models have you found useful through Ollama?
- I was never able to get these models to collaborate with me the way Opus does. I'm probably an outliner, I don't one-shot projects, I don't vibe code. I basically use LLMs are if I was working with a coworker, fairly smart one, but with short memory and often missing the big picture. Sometimes I can delegate more, sometimes less, but I know I always have to stay on top of what's happening, because it WILL create mess when it hits something hard. With the Antropic models, this kind of cooperation is easy (with the exception of Opus 4.6, which was bad for some reason).
- What do they do instead of collaborating?by therealdrag0
- A lot of open weight models don't understand intent well, they'll overfixate on a word in the prompt or just go off the rails trying to do much work.
GLM-5.2 actually has really good intent understanding though, on par with GPT-5.5 and Opus from my experience.
by x312 - > Opus 4.6 which was bad for some reason
If I recall, that model had a couple issues. One was the issue of being monkeyed with, for which they gave everyone credits.
The other feature/bug, depending on your POV, was being Anthropic's least personable release, not papering over everything with self help guru therapy language.
Opus 4.6 didn't LARP. It was more direct, less fussy, less discussy, and very much less "wait, one more thing" within a couple edits after embarking on what should have been the spec, than 4.7 or 4.8 are.
When in engineer brain mode, working as as you describe (good old fashioned XP-style staff engineer pair programming with a language-savvy mentee not yet full-stack or system wise), I found the clearer I was about my goal and the better I could express it, the more often I'd get an expanded clarified response I could then iterate to steer for ever tighter cleaner more specified responses, then let it go build the whole thing without it agonizing and waffling.
The next two releases regressed on that dimension, wanting to figuratively "sit with" every decision and re-validate spiritual alignment along the way, no matter how clearly expressed.
Curiously to me, Fable seemed to hit the best of both worlds, I had the highest commit per turn with Fable, approaching 73%, where I'm usually under 17% of LOC written being good enough to commit, usually taking 9 - 11 turns to get the code where I'm comfortable with it.
Thanks to this, Fable cost more, but actually cost less, if that makes sense.
Arguably, Fable, and 4.6, played more outcome-correctness oriented than journey-experience oriented. It's easy to see how this could happen with human reinforced learning if not all judges are staff or principal engineer level, or constitution values are more Portlandia than Finlandia.
ANTHROP\C needs to balance these at the constitution level:
“We will work in a humane and thoughtful way, but production is the final judge. We will listen to people, but we will not let discussion replace decision. We will value craft, but not at the expense of usefulness. We will move fast, but not by hiding risk. We will measure outcomes, but not pretend that everything important is easy to measure.”
by Terretta - I feel like another comparison worth looking at is purely cost.
Capability per dollar is something I care about:
So you're really getting near opus level capability for the price of haiku.Opus API $5/$25 Sonnet API $5/$15 Haiku API $1/$5 GLM 5.2 API $1.4/$4.4 - Not really, GLM uses more tokens to get work done.
- I've been checking out GLM 5.2 on some projects and few thoughts on it:
- it takes it sweet time to get code rolling, not the fastest model by any means
- it strays a lot during discovery/planning but then corrects
- it's not steering friendly, as it hallucinates things that it doesn't follow later on
- its output is quite good
A sample use case: I was optimizing rendering on Swift+Zig codebase. It chocked on 5k data entries.
GLM 5.2 spent 20 minutes building the benchmarks and getting data out, which made me frustrated so I blocked non-editing tool access and went AFK, after approx. 30 minutes I found that it used already-made benchmarks and some "conclusions" to optimize 3 choke points. Output pointed that it couldn't validate suspicions and asked for more data.
Implementation worked well, it was idiomatic and non-intrusive. I would even say that it was more idiomatic than GPT 5.5 effects on same repo.
I would opt in in using it more BUT GPT usually completes same requests 5x faster.
GLM 5.2 was spark for preparing and running inside isolated containers with JJ workspaces (so that multiple can be ran in parallel).
by xlii - This mirrors my experience. I have been using it in Pi. It is smart and output is good but it is not efficient in getting there.by Imanari
- Also pricing, I wanted to give a try, but when pricing is only 30% cheaper than Opus, I wouldn't go for it with these issues.by Oras
- >it takes it sweet time to get code rolling, not the fastest model by any means
Which provider are you using? I got a z.ai Lite Coding Plan and it's my understanding z.ai is on the slower side of providers and the Lite plan gets lower priority on top of that. In the api key console, it shows dipping below 60 tok/sec which is quite slow.
by nijave - I used it the other day for something of low importance that other models simply weren't figuring out and I didn't want to burn up Opus 4.8 on. (It had to do with overriding left-click on a macOS menu bar and then making Ctrl+click or right click bring up the menu like left-click normally does, and doing all this conditionally.)
Switched the model to GLM-5.2 halfway in the middle of a troubleshooting session (didn't even bother to reprompt, just changed it in the middle of its reasoning), gave it a few minutes, problem fixed. This is with the subscription based allocation on OpenCode Go, where a problem like this would completely burn up my Opus for the current 5 hours or even the current week.
by trollbridge - Its also nice that you can see its entire reasoning trace. I can see it going off the rails - or see something I forgot to tell it - and stop and correct it. Or I'll learn WHY it made the choice it did and not have to question it after.by jeremyjh
- At work we use Anthropic models and have basically no limits. So I am very familiar with what Opus can do. I also see the bills, I know what it costs.
At home I make a point of trying other models / tools on my side projects. So I've been using OpenCode and trying tons of models via OpenRouter. I tried Kimi, Deepseek, MiMo, etc.
GLM 5.2 is a _major_ step up from every other non-GPT/Claude/Gemini model I've tried. It's not as good as latest Claude Opus, but it feels every bit as good as Opus from ~4 months ago at a fraction of the price.
To me this model is the "it just works" moment for open weights models. We had this for closed weights models in late 2025 when Opus 4.5 landed. This is the same feeling I'm having with GLM 5.2. It's 90% as good as what I get from Anthropic for 1/5th of the cost and without any concern of lock-in.
by habosa