GLM 5.2

Author states 'The best open model in the world right now isn't from a company called OpenAI. It's of course from a Chinese lab. And this one is GLM 5.2 from ZAI.'

Author says 'matching GPT 5.5 on certain benchmarks.'

Author says 'there's even a category where it appears to be beating Fable all while being MIT licensed open weight.'

Author says 'it's very impressive that they made such a leap... GLM 5.2 here got a score of 51, which is 11 ahead of its previous iteration.'

Author lists next models and says 'top open model by a pretty healthy margin.'

Author says 'places it in the same realm as Gemini 3.5 Flash and GPT 5.4 on a max effort, which is pretty insane.'

Author says 'on a few of the benchmarks... like GPD Valve, actually outscores GPT 5.5.'

Author says 'it actually outscores Opus 4.7 on a medium effort... It is worth noting... not every single model has been tested... harness used was actually clawed code.'

Factual statement about license.

Presented as a specification.

Specification detail.

Author states 'on the coding index, it scores the same as Gemini 3.1 Pro and actually beats Sonic 4.6, and it isn't even that far off the top frontier models.'

Author says 'It's also a fair bit ahead of Kimmy K 2.7 Code, which is their newest model.'

Author says 'GLM 5.2 just took first place overall... becoming the first model ever to beat the Claude line, including Fable 5.'

Author cites Design Arena findings: 'a strong set of expert templates that avoid common AI antiatterns, so you should get less purple gradients... works really well with common libraries like ChartJS, 3JS, and Tailwind.'

Author states 'It sits second on GameDev, data viz, and 3D and fourth when it comes to UI components.'

Author says 'It does come with a little trade-off that it's a bit slower, but I'll come back to that later.'

Author says 'When it comes to speed, GLM 5.2 is actually not bad at all. It outperformed most of the open models... and it's a bit behind a Frontier model like Gemini 3.1 Pro.'

Author says 'Design Arena... say that GLM 5.2 scores the highest on user preference... but it was also the slowest out of the top models... all of those top models are Frontier ones and not open ones.'

Author notes as an annoyance: 'it only accepts text modalities. So you can't upload a screenshot and say recreate this.'

Author says 'the results I got back were super impressive... got the overall elements right... recreated the UI, which I think was very cool.'

Author says 'I'm not sure that I can agree with Design Arena that this doesn't have the usual AI look. This is really using those purple gradients to the max.'

Author says 'this one got to work... took overall about 10 minutes. ...used 40,000 tokens and cost 32.' Then describes controls issues.

Author compares models: 'Claude Opus 4.8... gave us the most playable demo in a single prompt'. Kimmy needed error fixing.

Author: 'Here's GLM 5.2's attempt... everything appears to be working... did a very good job from that single prompt... went with a Nex.js application and it used Prisma for the database.'

Author compares: Kimmy gave almost same app but missing extras; Opus used in-memory store. He says 'I actually think GLM 5.2 may have won this one.'

Author states 'I think for a lot of tasks, you could secretly swap GLM 5.2 in the place of Sonet or even Opus for simpler tasks and I probably wouldn't notice.'

Author says 'It's one of the first open models that I haven't felt like I'm fighting to use and also one of the first open models where using it I haven't had that feeling of I know Claude could do this better or faster.'

Author says 'I do have to agree with that sentiment where actually using these models feels a little bit different. But I think GLM 5.2 is one of the first ones that's broken that cycle for me.'

Author says 'If you told me a year ago that these open models would be anywhere near this good, I would have been absolutely shocked and probably not believed you.'

Author says 'One of the downsides... it's a little more token hungry when compared to other models... used an average of 43,000 tokens a task, which is more than...'

Presented as factual price information: 'it's around $140 for a million input tokens and $440 for a million output tokens. ...it actually cost around 50 cents a task.'

Author says 'at its intelligence level, GLM 5.2 is the cheapest model.'

Author states 'if you can take a hit to the intelligence, I do think Miniax and especially Deep Seek V4 are very good for that price.'

Author says 'it really feels like we're at a point where these open models are, let's say, 4 to 6 months behind... we could be looking at a fable model by next year. And I mean, they themselves are actually promising by Q1.'