China’s fast-growing AI scene had a big moment this week as Beijing-based startup Z.ai, formerly known as Zhipu, rolled out its newest open-source model, GLM-4.5.
The company is aiming to make waves by undercutting rival DeepSeek, which has already grabbed attention with its low-cost large language models.
Z.ai is betting that by offering an even more affordable and accessible tool for developers, it can carve out a bigger slice of the global AI market.
How GLM-4.5 stands out
Copy link to section
Z.ai’s new model, GLM-4.5, is being pitched as an “agentic AI,” basically, it can break down complex tasks into smaller, more manageable steps on its own, which should make it more efficient and useful in real-world applications.
Like DeepSeek, GLM-4.5 is open-source, so developers can download and use it for free, a big deal in the ongoing push to make advanced AI tools more accessible worldwide.
What really sets GLM-4.5 apart, though, is its pricing.
Z.ai is offering usage at just $0.11 per million input tokens, a bit cheaper than DeepSeek’s R1 model, which runs at $0.14.
Output token costs are also lower, making it one of the most budget-friendly options on the market.
That kind of pricing could be a game-changer, especially for smaller teams, research labs, and startups that need powerful AI tools but don’t have big tech budgets.
‘Intentionally lightweight’
Copy link to section
Z.ai’s CEO, Zhang Peng, says their new GLM-4.5 model is intentionally lightweight, about half the size of DeepSeek’s V3, and can run smoothly on just eight Nvidia H20 chips.
Those chips were built specifically for the Chinese market to stay within the boundaries of US export controls.
According to Zhang, they’re not facing any chip shortages at the moment, so supply isn’t a pressing issue, at least for now.
The rollout of GLM-4.5 comes at a time when US sanctions have forced Chinese tech companies to rethink their reliance on Western hardware.
There’s been a noticeable push to build up a more self-reliant AI ecosystem, with local alliances and industry groups racing to develop homegrown tools and infrastructure.
For firms like Z.ai, this isn’t just about innovation; it’s also about staying in the game without depending too heavily on tech that could be taken off the table at any time.
Fierce competition in China’s AI industry
Copy link to section
Z.ai’s launch comes on the heels of a bigger wave of AI innovation coming out of China.
Ever since DeepSeek’s R1 and V3 models made headlines for delivering solid performance at a fraction of what it costs to train Western models, other major players, including Alibaba, Tencent, Moonshot, and Baidu, have jumped in with their own models or upgrades.
The common thread? Keep costs low and make these tools widely accessible, both at home and abroad.
That strategy seems to be working. DeepSeek’s V3, for example, shot to the top of the US download charts shortly after launch.
Meanwhile, Moonshot’s Kimi K2, another open-weight model from a Chinese startup, is drawing attention for claiming better performance than some Western rivals in certain areas, all while offering cheaper API access and fewer restrictions for researchers.