Low-cost Chinese AI model closes in on Anthropic and OpenAI

A new low-cost Chinese AI model is drawing attention in the West, with experts saying it is closing the performance gap with leading US systems from OpenAI and Anthropic.

Since DeepSeek shook the AI industry early last year with a powerful yet inexpensive model, users have largely faced a trade-off between affordable Chinese models with relatively limited capabilities and more advanced but costlier offerings from OpenAI and Anthropic, which have invested billions in AI development.

That balance may be shifting with GLM-5.2, launched last month by Beijing-based startup Z.ai. The model has gained traction in Silicon Valley for its strong coding skills and agentic capabilities—the ability to complete complex tasks with minimal user input—delivering performance close to leading US models at a fraction of the cost. Some industry observers have described its emergence as a "mini DeepSeek moment."

GLM-5.2 has rapidly climbed the rankings on AI developer platform OpenRouter, overtaking Anthropic's models in usage. It has also earned praise from prominent technology leaders, including Snowflake CEO Sridhar Ramaswamy and venture capitalist Marc Andreessen.

"We now have a Chinese open-weight model that is as good as the currently available models from OpenAI and Anthropic," said David Sacks, former AI adviser to US President Donald Trump, shortly before Washington lifted restrictions on Anthropic's Fable and Mythos models earlier this week.

The model has intensified debate over whether China is beginning to narrow the AI gap with the United States, as some technology executives warn that regulatory uncertainty in Washington could weaken America's competitive edge.

Speaking on the All-In podcast, Sacks said GLM-5.2 performs just below Anthropic's Opus 4.8 model and is comparable with OpenAI's GPT-5.5, adding that the US "cannot afford to do things that slow our companies down."

Industry analysts say restrictions on Anthropic's models and the delayed public release of OpenAI's GPT-5.6 have boosted global interest in GLM-5.2.

Brian Tse, founder and CEO of Beijing-based AI consultancy Concordia AI, said developers are becoming increasingly wary of relying exclusively on proprietary US AI models because of access and policy risks.

GLM-5.2's popularity also reflects growing demand for affordable open-source AI, as businesses grapple with the rising costs of proprietary agentic AI systems, which consume large numbers of tokens during operation.

Z.ai, also known as Zhipu AI, declined to comment. OpenAI and Anthropic did not immediately respond to requests for comment.

According to Artificial Analysis, GLM-5.2 ranks fifth among leading large language models based on reasoning, coding and overall performance. It also holds second place on Code Arena's front-end coding leaderboard while operating at roughly one-sixth of the cost of frontier closed-source models such as Claude and the GPT series.

The company has not disclosed how much it spent developing the model.

Last month, Z.ai founder Tang Jie said on X that the company expects to build a model comparable to Anthropic's Fable before the end of the first quarter next year.

Former Hugging Face APAC lead Tiezhen Wang said GLM-5.2 stands out because it is essentially ready for deployment without requiring extensive fine-tuning, significantly lowering barriers to open-source AI adoption.

Data security remains a key obstacle

Despite its technical strengths, analysts say data security concerns could limit widespread adoption of Chinese AI models by US companies, particularly in heavily regulated sectors such as finance and cybersecurity.

Wang noted that migrating enterprise AI systems typically takes several months, while Counterpoint Research analyst Wei Sun said many clients in Europe and the US may remain reluctant to integrate Chinese AI models into their technology stacks regardless of their price or performance.

A RAND report released earlier this year found that Chinese large language models increased their global market share from 3% to 13% in the two months following DeepSeek's R1 launch. Adoption grew most rapidly in developing economies and countries with close political and economic ties to Beijing.

However, some experts argue that security concerns are overstated, noting that organisations can run Chinese open-source models on their own infrastructure or through US cloud providers without exposing sensitive data.

They also suggest that startups and small businesses are adopting these models much faster than large corporations.

"Developers tend to care less about where a model comes from than whether it works, how much it costs and whether they can deploy or access it reliably," said technology analyst Poe Zhao.

He added that GLM-5.2 is unlikely to replace OpenAI or Anthropic outright but instead will encourage businesses to use multiple AI models depending on their needs.

"The likely pattern is partial routing, not overnight replacement of OpenAI or Anthropic. So yes, it is a mini DeepSeek moment, but in a narrower, developer-centric sense."