Chinese AI model GLM-5.2 gains attention as lower-cost rival to US systems

BEIJING: A new artificial intelligence model from Chinese startup Z.ai is drawing increasing attention in global technology circles, with some executives and analysts saying it is narrowing the gap with leading US systems while offering much lower costs.

Z.ai, also known as Zhipu AI, launched GLM-5.2 last month. The model has attracted notice in Silicon Valley for its coding performance and agent capabilities, referring to its ability to carry out complex tasks with limited prompting. Some experts have described the development as a mini DeepSeek moment.

Interest in Chinese AI products had previously largely been framed as a trade-off between lower prices and weaker performance compared with products from OpenAI and Anthropic, which have invested billions in development. GLM-5.2 may be altering that perception in Western markets.

On third-party AI developer platform OpenRouter, GLM-5.2 has risen rapidly in usage rankings and now sits ahead of Anthropic’s models. It has also received praise from several prominent figures, including Snowflake chief executive Sridhar Ramaswamy and venture capitalist Marc Andreessen.

David Sacks, former AI czar to US President Donald Trump, said last week that the model was performing at a level comparable to leading US offerings. He made the remarks before Washington lifted curbs on Anthropic’s Fable and Mythos models on Tuesday.

Speaking on the All-In podcast, Sacks said:

"We now have a Chinese open-weight model that is as good as the currently available models ​from OpenAI and Anthropic"

He also said:

"It is just a tick below Opus 4.8 (from Anthropic) and right up there with GPT 5.5 (from OpenAI)"

Sacks added:

"we cannot afford to do things that slow our companies down."

Restrictions previously placed on Anthropic’s models and the delayed public release of OpenAI’s GPT-5.6 have added to international demand for the Chinese model. Brian Tse, founder and chief executive of Beijing-based AI safety consultancy Concordia AI, said developers were becoming more aware of the risks of relying only on proprietary US-based application programming interface models.

GLM-5.2 is currently ranked fifth on Artificial Analysis’ large language model intelligence leaderboard, which measures areas such as reasoning and coding. It is also in second place on Code Arena’s front-end coding rankings, which assess how well models build websites and front-end applications. The model operates at roughly one-sixth of the cost of closed US frontier models such as Claude and the GPT series.

Z.ai has not disclosed how much it spent to build GLM-5.2 and declined to comment. Anthropic and OpenAI did not immediately respond to requests for comment.

Last month, in a reply to Elon Musk on X, Z.ai founder Tang Jie said the company could deliver a model on par with Anthropic’s Fable before the first quarter of next year.

Tiezhen Wang, former APAC lead at Hugging Face, said the model’s significance lay in making open-source AI much easier to use. He said:

"The shift GLM-5.2 brings is that the open-source model has become a plug-and-play, out-of-the-box product"

He added:

You just deploy the model and without doing any complex fine-tuning systems, it is in a highly usable, ready-to-use state. This drastically lowers the barrier to entry for open-source adoption.

Security concerns and adoption

One of the main barriers to wider enterprise use, especially in the United States, is concern over data security. That issue has limited adoption of Chinese AI systems in regulated sectors such as banking and cybersecurity.

Wang said migration and upgrading of enterprise AI systems typically takes several months. Wei Sun, principal AI analyst at Counterpoint Research, said some European companies were discussing whether the model could be used in enterprise settings, but added that clients, partners and regulated industries in the European Union and the United States might still reject Chinese models regardless of price or technical quality.

A RAND report published earlier this year based on website traffic data from 135 countries found that Chinese large language models increased their global market share to 13% from 3% in the two months after DeepSeek released its R1 model in January last year. The gains were strongest in developing countries and in states with close political and economic ties to Beijing.

Some experts said fears about the safety of Chinese AI models were overstated, arguing that data could remain secure if the systems were run through US cloud providers or on a company’s own servers.

While large corporations tend to move slowly, startups and small- and medium-sized businesses are adopting new models more quickly. Poe Zhao, founder of the Hello China Tech newsletter and a China technology analyst, said developers are generally more concerned with performance, cost and reliable access than the country where a model originates. He said the likely outcome would be partial routing rather than a sudden replacement of OpenAI or Anthropic, and described the development as a narrower, developer-focused mini DeepSeek moment.