Budget Chinese AI Model Rivals Anthropic and OpenAI at Home

Select Language:

Since DeepSeek startled the market early last year with its affordable yet powerful AI model, consumers worldwide have faced a dilemma: opt for Chinese models with lower prices and fewer features or choose OpenAI or Anthropic, which have invested billions into refining their technologies.

Recently, Beijing-based startup Z.ai introduced GLM-5.2, a model launched last month that might be bridging that gap in Western interest. It’s generating buzz in Silicon Valley for its coding and agent functionalities—the ability to perform complicated tasks with minimal prompts—coming close to matching leading US models but at a fraction of the cost. Some analysts are calling it a “mini DeepSeek moment.”

GLM-5.2 is rapidly climbing usage charts on third-party AI development platforms like OpenRouter, now outranking models from Anthropic. Industry leaders, including Snowflake CEO Sridhar Ramaswamy and venture capitalist Marc Andreessen, have praised its performance.

“Now we have a Chinese open-weight model that rivals the current offerings from OpenAI and Anthropic,” declared David Sacks, former AI czar for President Donald Trump, last week, shortly before the US lifted restrictions on Anthropic’s Fable and Mythos models.

This shift has fueled debates about whether China is finally catching up to the US in AI leadership, with tech executives warning that Washington’s unpredictable regulations could threaten the nation’s dominance in frontier technology.

Sacks added on the All-In podcast, “GLM-5.2 is just below Anthropic’s Opus 4.8 and on par with GPT-5.5 from OpenAI. We simply can’t afford to slow down our companies’ progress.”

The restrictions on Anthropic and the delay in launching OpenAI’s latest GPT-5.6 have heightened worldwide interest in Chinese models. Industry insiders note that reliance solely on US-based proprietary API models carries considerable risk, prompting developers to explore open-source alternatives.

The positive reception of GLM-5.2 indicates a rising interest in cost-effective, open-source development, especially as businesses grapple with escalating expenses related to AI usage—costs that often fluctuate unpredictably because closed-source, agentic AI tools tend to consume more tokens, the units measuring AI activity.

Z.ai, also called Zhipu AI, declined to comment, and representatives from Anthropic and OpenAI did not immediately respond to requests for remark.

Currently, GLM-5.2 ranks fifth on Artificial Analysis’s large language model (LLM) performance leaderboard, which assesses capabilities across various benchmarks like reasoning and coding. It also holds second place in Code Arena’s front-end coding rankings, demonstrating proficiency in generating websites and applications at about one-sixth the cost of US models like Claude and GPT.

Z.ai has not disclosed how much it invested in developing GLM-5.2. In a reply on X (formerly Twitter) last month, founder Tang Jie stated the company expects to produce a comparable model to Anthropic’s Fable before the first quarter of next year.

“The key advantage with GLM-5.2 is that open-source models are now ready-to-deploy, plug-and-play solutions,” said Tiezhen Wang, a former APAC lead at Hugging Face—an organization serving developers working with open-source models. “You can deploy the model immediately without extensive fine-tuning, significantly lowering the barrier for adoption.”

However, widespread adoption faces hurdles, particularly around data security concerns that make US enterprises hesitant to use Chinese models, especially in regulated sectors like banking and cybersecurity. Transitioning and upgrading enterprise AI systems can take months, Wang noted.

Wei Sun, principal AI analyst at Counterpoint Research, mentioned, “Some European companies are also discussing whether these models are viable for their enterprise needs. Still, many in the US and EU—clients, partners, and regulated industries—are often unwilling to incorporate Chinese AI models into their stacks, regardless of their technical merits or cost advantages.”

A recent report by the nonprofit Rand Corporation analyzed web traffic data from 135 countries and found that Chinese LLM market share surged from 3% to 13% within two months after DeepSeek introduced its R1 model in January last year. This development caused a global tech selloff due to DeepSeek’s low-cost offerings contrasting with high infrastructure investments elsewhere.

The usage increases are most notable in developing nations and regions with close economic or political ties to Beijing. Many experts argue that safety concerns over Chinese AI models are exaggerated, contending that running these models on US cloud services or on private servers can safeguard data security. While larger corporations tend to stay cautious, startups and smaller firms are adopting these models more swiftly.

“Developers care less about the origin of a model than its functionality, cost, and reliability,” said Poe Zhao, a China tech analyst and founder of the Hello China Tech newsletter. “The most likely scenario isn’t an overnight replacement of OpenAI or Anthropic but a gradual, partial integration—what I’d call a mini DeepSeek moment, focused on developers.”