China's Cut-Price AI Model Is Closing In on Anthropic and OpenAI

.

A Cheap Alternative Gets Serious

For more than a year, buyers of artificial intelligence software have faced a trade-off. Chinese models were cheap but usually weaker. American models from Anthropic and OpenAI were powerful but expensive, after the companies spent billions of dollars building them.

That gap now looks smaller. GLM-5.2, released in mid-June by the Chinese startup Z.ai (also known as Zhipu AI), performs close to Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 on independent benchmark tests (standardized exams that measure how well an AI model reasons, codes, or follows instructions). On some measures it trails the best American systems by only a percentage point, while charging roughly a fifth to a tenth of the price per query.

The model has quickly become one of the most-used AI systems on OpenRouter, a marketplace where developers compare and pay for different AI models. Investor Marc Andreessen and Snowflake chief executive Sridhar Ramaswamy have both praised its abilities publicly.

.

Why the Timing Matters

GLM-5.2's rise did not happen in a vacuum. On June 12, the U.S. Department of Commerce ordered Anthropic to suspend global access to its newest models, Fable 5 and Mythos 5, citing national security concerns tied to their cybersecurity capabilities. Because the order applied to all foreign nationals, Anthropic switched the models off worldwide rather than risk violating it.

Z.ai released GLM-5.2's open weights (the underlying files that let anyone download and run a model on their own computers) just days later, on June 16. OpenAI, facing a separate government request, also delayed the full rollout of its GPT-5.6 model that same week.

David Sacks, a former White House AI adviser under President Trump, called GLM-5.2 "a tick below Opus 4.8" and "right up there with GPT-5.5" in comments on the All-In podcast. He argued that unpredictable U.S. regulation risks slowing American companies down at a critical moment in the AI race.

The dispute was resolved on June 30, when Commerce lifted the export controls after Anthropic added a new safety filter. Anthropic restored global access to Fable 5 on July 1. But the 18-day gap gave Z.ai a rare window to win over developers who might otherwise have stayed with U.S. providers.

.

What Makes GLM-5.2 Different

GLM-5.2 is a mixture-of-experts model (an architecture that activates only part of a very large network for each task, making it cheaper to run) with roughly 744 billion total parameters, of which about 40 billion are active at any one time. It supports a context window of one million tokens, meaning it can process very long documents or codebases in a single request.

Unlike Anthropic's or OpenAI's flagship systems, GLM-5.2 is released under an MIT license. That means any company or developer can download it, modify it, and run it on their own servers indefinitely — without needing permission, and without the risk that a government order could switch it off overnight.

Independent evaluators have taken notice. Analysis firm Vals AI found GLM-5.2 trailing Anthropic's Opus 4.5 by just one percentage point on a reasoning test called ProofBench. On Artificial Analysis' broader intelligence ranking, GLM-5.2 currently sits in the top five worldwide.

.

The Limits of the Comparison

The picture is not one-sided. GLM-5.2 still trails the leading Anthropic and OpenAI models on some technical benchmarks, including terminal-based coding tasks, where it scored several points behind Opus 4.8. Developers testing the model in production have also raised questions about the transparency of how Z.ai reports token usage and costs.

Security is a bigger concern for potential Western customers. Z.ai, like all companies operating in China, is subject to Chinese national security law, which can compel firms to share data with state authorities on request. U.S. lawmakers opened a formal inquiry in May into the cybersecurity risks posed by Chinese-origin AI models in critical infrastructure.

Analysts note an important distinction, however: once a company downloads GLM-5.2's open weights and runs the model on its own servers, no outside party — including Z.ai itself — has ongoing access to that system. The greater risk applies mainly to companies that use Z.ai's cloud-hosted API rather than self-hosting the model.

China's growing AI capability also unfolds against a backdrop that will be familiar to Falun Gong practitioners and human rights observers: the same Chinese Communist Party that censors information domestically and has for decades persecuted Falun Gong is now positioning its technology companies as global standard-setters in artificial intelligence, a technology increasingly used for surveillance and information control at home.

.

Where This Leaves the Market

For now, most analysts describe the shift as gradual rather than sudden. "The likely pattern is partial routing, not overnight replacement of OpenAI or Anthropic," China tech analyst Poe Zhao said, describing the moment as a narrower, developer-focused echo of the reaction sparked by DeepSeek's R1 model in early 2025.

Large regulated companies in banking and cybersecurity are expected to remain cautious about Chinese models regardless of price, given compliance requirements. But smaller companies and startups — more sensitive to cost and less bound by regulation — are already testing GLM-5.2 and other Chinese open-source models, and some have begun shifting workloads away from U.S. providers entirely.

The episode highlights a deeper vulnerability for American AI companies: as long as access to their most advanced models can be interrupted by regulatory decisions, cheaper and permanently downloadable open-source alternatives from China gain a structural advantage — regardless of how the underlying technology compares in a lab test.

.

Sources

1. CNBC – "China's Zhipu is closing in on top U.S. AI models with Anthropic and OpenAI held back": https://www.cnbc.com/2026/06/26/china-zhipu-z-ai-open-source-anthropic-openai.html
2. CNBC – "White House AI crackdown opens door for Chinese model makers to close gap": https://www.cnbc.com/amp/2026/06/30/white-house-ai-china-crackdown.html
3. Tom's Hardware – "Chinese Z.ai's latest model tops AI ranking charts amid Anthropic Fable 5 ban": https://www.tomshardware.com/tech-industry/artificial-intelligence/z-ai-free-glm-5-2-tops-the-open-weight-ai-rankings-on-all-huawei-silicon
4. Euronews – "US lifts export controls on powerful AI models, Anthropic says": https://www.euronews.com/2026/07/01/us-lifts-export-controls-on-powerful-ai-models-anthropic-says

.