The new contender is Chinese.
And it is fast. Z.ai just dropped GLM 5.2 into the fray, catching up to the heavyweights from the West with a speed that feels almost aggressive. They launched this beast a day after the US slapped an embargo on Anthropic, banning Fable 5 and Mythous models for non-Americans. The restrictions vanished on June 30, but the message stayed clear. Timing matters.
Z.ai claims their model sits right beside the giants. We are talking about parity with Anthropic’s Claude Op 4.8 and Openai’s GPT-5.5. The context window? Massive. One million tokens. Roughly 750,00 words of working memory held at once.
It handles code. Messy, long, complex code.
According to the makers, GLM-5 maintains quality even when the coding-agent trajectory gets messy and stretched out. They put it through the wringer on three benchmarks involving long, heavy coding tasks.
- Open-ended technical projects. These can last hours, sometimes days. Here, GLM-5 lags Opus 4.8 by a razor-thin 1%. It edges past both GPT-5 and the earlier Opus 4.
- Model improvement. Tested on improving smaller models with just one GPU, it beat GPT-5 and Opus-4. Second place overall. Right behind the big guy.
- Engineering marathons. Think compilers. Heavy lifting. Z.ai says it trails Opus-4 by 13%. A noticeable gap, sure. But still the second-best overall.
Across the board, though? They call it the leading open source model.
The Open vs. Closed War
This is the real story. Not just speed, but freedom.
GLM-5 is open source. Z.ai notes there are “no regional limits” and “technical access without borders.” That is a mouthful, but it means one thing. You can take it. Break it. Modify it. Change its output. Share it with or without alterations, for any reason you want.
Anthropic? Closed.
OpenAI? Closed.
With their models, you depend on the provider. You cannot touch the weights. You cannot see inside the box.
“This means that the AI system can be adapted for any purpose…”
That adaptability is the weapon here. The US and China are racing. Not just for tech supremacy, but for control over the infrastructure of future healthcare, national security, and everything else that will rely on silicon intelligence. The US strategy has been restriction. They block semiconductor access, try to choke off the supply chain, get ahead by closing the door.
China? They are building a different door.
One made of open-source code. Cheaper. Wider.
Remember January last year? DeepSeek released R1. It was cheaper. More energy efficient. It rattled the cage of US competitors before. GLM-5 feels like the next step in that same playbook.
Is the west worried about open access? Probably. Should they be?
The gap narrows when you can run the code locally. When you can tweak the engine without asking permission. GLM-5 sits at number two on some charts. Number one on openness.
And that distinction might end up being the difference.
