NEW Chinese AI DESTROYS Mythos?

**Julian Goldie** (0:00)
China's Glm 5.2 beats or matches Anthropic's Mythos. So we're looking at a situation that just got released today. It actually states that the band model, Mythos and Fable 5, has already been matched by China's ZAI. So they own Glm 5.2, and this was announced by the Wall Street Journal. So according to the Wall Street Journal, China has matched Anthropic in cybersecurity, resetting the AI race. Now, there's so much more to it, which I'll come on to in a second, but this is matching in certain benchmarks, such as exploiting cybersecurity bugs, which is pretty crazy stuff when you think about it, because bear in mind, this is a fully gated model. Nobody can access Fable 5 apart from a small group of cybersecurity firms. But we're looking at China at this point, with an open model ZAI and Glm 5.2 matching Claude Mythos at discovering software vulnerabilities, making a huge leap in the Chinese and US frontier AI race. So apparently, according to this, as you can see this tweet by the Bureau, China's ZAI can match Mythos, and it achieves Mythos level vulnerability detection. Now, when I was testing out Glm 5.2, it certainly scored better than Opus 4.8 on Goldie Bench. Really powerful model, but also this is number one, Chinese. And number two, it is an open source model, which means, you know, if you've got a great setup, number one, the weights are open and shared and hugging face. But number two, you can host it locally if you have a great setup. And so China have matched Mythos. And this is something that I've always been saying, especially recently, is like the model is never the most. An open Chinese model just matched Anthropic's most locked down system at finding security bugs. And it's obviously for like a quarter of the cost if that. So the Wall Street Journal called it a reset of the AI race. And here's what you can learn from this. So we've got Claude Mythos, which is fully gated, never actually released publicly. This is the cybersecurity version. And you've got Fable 5 that was released, but then taken down by the US government. And they basically said, nobody can access to this right now.
And it's been released to a tiny small group of people. Maybe it will come back, but there's no guarantees. Then you've got Glm 5.2 from ZAI, Chinese, open way, free to download, matched a quarter of the cost if you're using the API. And the open model anyone can download just matched the most locked down frontier model at roughly a quarter of the cost per token. Now this story was just announced a few hours ago, it's brand new, this is from the Wall Street Journal, as you can see, so China's matched Anthropic. And this is creating really an advantage for China, they're catching up. I'm probably gonna overtake, because bear in mind like Glm 5.2 as well, is a self-learning model, a self-iterating model, which means Glm 5.3 will be a big step up again. And it only gets better from here, like Chinese AI is the worst it's ever gonna be. So if we look at the news, you know, Mythos is Anthropic's most powerful model, it's good at finding security holes in software, particularly, in fact, so good that they keep it locked away and only a vetted handful of US agencies and companies can touch it. And this week, China's ZAI said its model now does the same job, finding the same security bugs, matching the most guarded model in the world. And its model, Glm 5.2, is open. Anyone can download it and they can use it for about a quarter of the cost per token. If you're using the coding plan with Glm 5.2, you can plug it into any of your agents and it's very affordable. I've never run out of tokens on that model. So pretty wild stuff. Also, something that's interesting that Anthropic have come out with recently is that Chinese models are basically distilling and copying the models. And there's basically rumors of China backdooring the model.
And that was floating around even before Fable 5 shipped. So whether you believe that or not, it's pretty interesting stuff. This whole race is really heating up right now. There's also a different situation going on. So China didn't just like, you know, they didn't copy a model. They've actually built their own platform. So a second Chinese firm called 360 Security has shipped its own cyber AI the same week, making the same claim that they're beating Mythos. And they're building like a whole stack. So this is like a homegrown rival to Mythos. And they've actually got receipts of that. So there's one model that auto discovers software vulnerabilities and a second system that runs cyber defense.