Chinese AI startup Z.ai targets AGI as GLM-5.2 beats GPT-5.5 and narrows the gap with OpenAI and Anthropic

Just a week after China’s Z.ai grabbed global attention after its GLM-5.2 outperformed GPT-5.5 on coding benchmarks at a fraction of the cost, the company said Thursday it plans to use proceeds from a domestic listing to fund its push toward artificial general intelligence (AGI).

Z.ai, the Chinese AI startup formerly known as Zhipu AI, said Thursday that proceeds from a planned domestic listing will go toward advancing its models toward AGI. The company’s latest model, GLM-5.2, has quickly become a breakout moment for China’s open-source AI push after landing near the top of public leaderboards dominated by Anthropic and OpenAI.

“Our mission is to obtain AGI, so right now our focus is on how to improve our model to achieve the upper bound of intelligence. So all these resources are helping us,” said Qinkai Zheng, technical lead of the firm’s CodeGeeX team.

The timing matters. Z.ai released GLM-5.2 a day after Anthropic cut off worldwide access to its most advanced models, a move that rattled developers and governments already uneasy about relying too heavily on U.S.-controlled AI infrastructure. In that vacuum, Z.ai’s latest release landed with unusual force. The model now ranks fourth on Artificial Analysis’ LLM intelligence leaderboard and second on Code Arena’s front-end coding leaderboard, according to the company, with operating costs roughly one-sixth of leading closed U.S. frontier models.

That combination of performance, lower cost, and open-source availability has turned Z.ai into one of the most closely watched AI companies in China. Its shares have surged more than 2,000% since the company’s Hong Kong debut in January, lifting its market capitalization past HK$1 trillion, or about $128 billion, this week.

“This model is comparable to the top closed models,” Zheng told reporters at the company’s Beijing headquarters, according to Reuters. “It’s the first time that an open-source model really delivers very solid coding and agent performance that can compare with the leading proprietary AI companies like Anthropic and OpenAI.”

For years, Chinese AI labs have been viewed as trailing top U.S. players by several months, especially in frontier model performance. Z.ai’s rise does not erase that gap overnight. It does suggest that at least one Chinese open-source lab is moving closer to the front line than many expected. The company had already surpassed open-source U.S. models such as Google’s Gemma and Meta’s Llama series on some benchmarks. GLM-5.2 appears to push that progress much further, bringing it closer to closed models from U.S. leaders that were once seen as comfortably out of reach.

Z.ai said earlier this month that it plans a dual listing in Shanghai, though it has not disclosed how much it hopes to raise. The money would support a strategy centered on bigger and more capable models, with a particular focus on coding, complex multi-step tasks, and autonomous agents that can work through long-horizon problems with less human intervention.

Z.ai Races Toward AGI After Chinese AI Startup’s GLM-5.2 Challenges OpenAI and Anthropic

GLM-5.2 is built for that kind of work. The model has 750 billion total parameters and a 1 million-token context window, giving it room to process far larger volumes of information than standard consumer chatbots. Z.ai said it released the model with inference support for a broad range of domestic chip infrastructure, including Huawei Ascend clusters, a notable detail at a time when Washington has tightened restrictions on China’s access to advanced Nvidia hardware.

Since February, the GLM-5 series has been adapted to run on Chinese semiconductors after new U.S. export controls narrowed access to the most advanced AI chips. Zheng declined to say whether GLM-5.2 was trained on domestic or foreign hardware, but said the company is working to improve efficiency across multiple chip architectures.

“We are trying our best to improve our infrastructure and to make the model more efficient on different kinds of chips,” he said.

That chip flexibility could become one of Z.ai’s biggest advantages inside China, where AI labs are under pressure to build around hardware constraints rather than assume access to Nvidia’s latest systems. It could help explain why Z.ai has gained traction with enterprise customers and public-sector buyers at a time when China’s AI market has been locked in a bruising price war.

Even in that environment, the company has raised prices for its frontier models several times this year, a sign that demand for its higher-end systems may be strong enough to give it some pricing power. Zheng suggested more increases could come if usage keeps climbing.

“We are trying to lower the cost, but because the demand is too large, maybe in the future we will still need to increase the price,” he said. “But we want the model accessible to everyone.”

Investors are paying attention. JPMorgan has projected that Z.ai’s revenue could jump more than 534% this year, with profitability expected in 2028. That said, the company still generates only a small fraction of the revenue produced by leading U.S. AI firms, according to stock exchange filings. Z.ai is still early in the process of turning technical momentum into a business on the scale of OpenAI or Anthropic.

Its next test may come quickly. Zheng said the company is already working on GLM-5.5, expected in August, and future releases will focus on long-horizon tasks and self-evolving autonomous agent systems. If GLM-5.2 was the model that forced the market to take Z.ai seriously, GLM-5.5 may be the one that shows whether this was a one-off leap or the start of a more durable shift in the global AI race.