Chinese AI lab Z.ai (formerly Zhipu AI) released GLM-5.2, an open-source model with 750 billion parameters and a 1-million-token context window, running entirely on domestic chips. Priced at roughly one-sixth of US frontier models, it scores within 1% of Anthropic's Opus 4.8 on agentic benchmarks and shows a 2.3x improvement over its predecessor in app development tasks. The launch came a day after Anthropic disabled global access to its advanced models, and amid US export restrictions. Z.ai co-founder Tang Jie criticized the access suspensions as 'deeply regrettable,' positioning open-source models as a more reliable choice. Markets reacted positively, with Z.ai shares surging over 30% in Hong Kong. The cost advantage is significant: DeepSeek V4 Pro charges $3.48 per million output tokens versus $50 for Anthropic's Fable 5. Chinese models now hold the top four slots on OpenRouter by token traffic, signaling a shift in enterprise adoption toward open-source, lower-cost alternatives.

Key facts

• GLM-5.2 has 750B parameters, 1M-token context, runs on domestic chips
• Priced at ~$3.48 vs $50 for Anthropic Fable 5 per million output tokens
• Scores within 1% of Opus 4.8 on agentic benchmarks
• Chinese models top OpenRouter traffic: DeepSeek, MiniMax, Tencent, Xiaomi
• Z.ai shares rose 30%+ after launch; revenue projected to grow 534% this year