China’s AI Models Are Catching Up. Z.ai’s GLM-5.3 Takes Aim at OpenAI & Anthropic

China's Z.ai is escalating its challenge to U.S. artificial intelligence leaders Anthropic and OpenAI with a new model designed to narrow the gap in AI-powered coding, one of the industry's most competitive areas.

The Beijing-based company said its upcoming GLM-5.3 model will deliver improved coding capabilities, putting its performance closer to leading systems such as Anthropic's Fable 5. Z.ai plans to release the model's weights within two weeks.

Making the underlying parameters available will allow developers to download, modify, and customize the system, part of the company's strategy to expand adoption through a permissive licensing approach.

The company released benchmark results indicating that GLM-5.3 significantly improves on GLM-5.2 and performs close to, and in some tests ahead of, Anthropic's Fable 5. The results are Z.ai's own benchmarks and will face additional scrutiny as independent developers and evaluators test the model after its wider release.

The announcement comes as competition between Chinese and American AI companies intensifies. Anthropic and OpenAI, both based in San Francisco, are widely considered leaders in frontier AI. But Chinese developers have increasingly demonstrated that their models can achieve comparable performance in some areas while operating at lower costs.

Models from Moonshot AI, DeepSeek and Alibaba have all performed strongly in artificial intelligence benchmarks. Moonshot's Kimi, DeepSeek's V4 models and Alibaba's Qwen have helped turn China's AI sector into a growing competitive threat to U.S. developers.

Independent evaluator Artificial Analysis currently assigns both DeepSeek's leading model and GLM-5.2 an intelligence score of 53. Both trail Moonshot's Kimi K3 and U.S. frontier systems including Anthropic's Claude Fable 5 and OpenAI's GPT 5.6, according to the report.

The economics of that competition are also shifting. DeepSeek announced Thursday that it would raise prices by as much as fourfold for its V4 Flash and Pro models. However, the blended price for V4 Pro remains below that of GLM-5.2, according to Artificial Analysis.

Investors were less enthusiastic Friday. Z.ai shares fell as much as 9% in Hong Kong, while rival MiniMax Group dropped 16%. Analysts attributed the declines to investors taking profits ahead of quarterly results following sharp gains and significant volatility across AI stocks.

The company, one of China's early large language model developers and the world's first LLM maker to go public, reached a market capitalization of $137 billion at its peak following the release of GLM-5.2. That temporarily put its valuation above Chinese internet companies including PDD Holdings and NetEase.

Its market value has since declined to about $80 billion, but the stock remains roughly 10 times higher than at the time of its Hong Kong listing in January. Z.ai is also expanding the infrastructure supporting its models.

The company has completed a data center equipped with at least 10,000 Chinese-made chips to develop and deploy its GLM systems, according to a Bloomberg report. Its annual recurring revenue reached $1 billion in July, reflecting growing spending by developers and businesses using its AI services.