Chinese AI labs are making their most aggressive push yet against US frontier models. Z.ai’s release of GLM-5.2 — an open-source model running at roughly one-sixth the cost of American alternatives — has shaken up the competitive landscape almost overnight.
The timing wasn’t accidental. GLM-5.2 launched just a day after Anthropic disabled global access to its most advanced models, including Fable 5 and Mythos. OpenAI also moved to limit access to GPT-5.6 following a government request the same week. Z.ai’s co-founder called the Anthropic suspension “deeply regrettable” and positioned Chinese open weights as the safer institutional bet.
GLM-5.2 packs 750 billion parameters with a 1-million-token context window, running entirely on domestic Chinese chips. Benchmarks show it within a single percentage point of Anthropic’s Opus 4.8 on agentic evaluations. The gap between Chinese open models and top US closed systems has shrunk faster than most forecasts predicted.
The cost advantage is stark. DeepSeek V4 Pro charges $3.48 per million output tokens. Anthropic’s Fable 5 charged $50 for the same output. Enterprise buyers are rethinking their AI vendor relationships as a result. OpenRouter data shows Chinese models now hold the top four positions among the most widely used systems globally by token traffic.
Z.ai’s shares surged over 30% in Hong Kong trading on the news, now up more than 800% since January. The company plans a dual listing in Shanghai and expects to launch GLM-5.5 in August. Whether the momentum holds depends on whether US labs restore access — and whether Chinese models can keep closing the capability gap.
