The Point of No Return: GLM 5.2 Approaches the Frontier

**Ejaaz** (0:00)
Last week, a Chinese company released a free AI model that is as good as Anthropic's best model. It also beats ChadGPD 5.5 at writing and coding, but it comes with a twist. It's a sixth of the price, and it's completely open source. You can download it and run it at home. Now, in that same week, the United States government banned Anthropic's most powerful model, Fable 5, after someone revealed that an unrestricted version of it had hacked into the National Security Agency's systems. I think we've reached a point of no return, and not to sound dramatic, but in six months, it is very realistic that we will have open source or open weight models that are accessible to anyone in the world with an internet connection and, I don't know, 5 to 10k to run at home, that they can fine tune to do anything. And it's METHOS grade level models. These are the same models that we're hearing rumors and reports from Verified that they can exploit some of the most secure systems in the world faster than any other exploiter has been able to do in the past. And I think we're going to look back on 2026 as the moment or the year that everything really changed and the point where humanity as itself really needs to focus on safeguards and figuring out how to regulate and release these AI models in the future. So we've reached a convergence of this really interesting trend where the most powerful models of the world are 3D available and open source available for anyone to access.
And the government, the United States specifically, has an off switch for their most powerful model. Yeah.

**Josh** (1:30)
It's been a couple of months, it seems, since we've had some news on the frontier of China. And you kind of forget about them every couple of weeks where they just kind of disappear, they quiet down. The new models come out, we see the fables, we see the mythos of the world. But then out of nowhere, they strike back. And seemingly every single time, it comes as a surprise at how powerful these new models have become. So to start with this, we have a new model from our favorite company to pronounce, Jipu, Chinese company. I feel like I want to name my dog, that is such a cute name.
But Jipu is doing something not so cute. They're actually releasing a model named Glm 5.2, which blew everyone's expectations out of the water. I remember way back, like six months ago when DeepSeq was doing this, where DeepSeq would release a model, everyone was like, wait, you did what? With what? And that's what this model feels like again. We're getting that moment again because this is an open weights model, which is not to be confused with open source, and we'll talk about that in a little bit. But this is an open weights model that is, if I'm correct about this, within one single point of the SwiBench Pro benchmark, which is the benchmark that a lot of people use for coding, of GPT 5.5, the frontier coding model from OpenAI. And that comes as a surprise because the cost, well, one, if you run it locally is free, but two, if you run it on a server is, like you said earlier, you just one sixth of the cost. So you're getting an incredible amount of coding capability for something that costs a fraction of what it costs if you were to go to one of these larger language models. And it seems to work almost as good, if I'm right. And this comes as a surprise to most people because every time we start to count China out, we're like, no, surely they can't catch up. They continue to chip away at this frontier.

**Ejaaz** (3:09)
There's a few things that people will jump to immediately. Okay, one, that these benchmarks can be easily gamed. We're going to show you a few examples of benchmarks that couldn't be gamed and Glm 5.2 performs really, really well. But the second thing is the cost. Cost has become a really important point of discussion amongst enterprises specifically that are spending hundreds of millions of dollars per year to access Claude and GPT. It's just too much money for them to spend in terms of the return on investment that they're getting in work that they actually see. What they're now turning towards is these free open-source models, primarily designed and made by Chinese AI labs, that can cut costs down drastically. Just last week, we had Microsoft announced that they're replacing their co-pilot LLM with not ChatGPT, with not Claude, but with DeepSeq itself. The point is, this comes at a very important time where cheaper models are getting a lot of attention. Now, when we look at Glm 5.2 specifically, it is five to seven times cheaper than GPT 5.5 and Claude Opus 4.8, but performs as we're seeing on the benchmarks right here, almost as good as each of these models, specifically at the metric that is the most important, which is coding. Now, a lot of skeptics, quite rightly, were like, I don't know if this is actually true. Let me test it against a few other independent benchmarks.