US export restrictions, particularly targeting Anthropic, have ignited a new phase of the AI race. We break down which models Asia is offering and what makes them unique.
Chinese company Z.ai has released GLM-5.2, an open-weight model with 750 billion parameters and a 1 million token context window. The model runs on domestically produced Chinese chips and lags behind Anthropic’s flagship Opus 4.8 by less than one percentage point on key benchmarks–at just one-sixth the cost.
Hot topic: Bitcoin Treasuries Buying Fails to Offset Selling, Glassnode Says
The release comes amid US export restrictions that have significantly narrowed foreign access to the most powerful American models, including Anthropic’s Fable 5 and Mythos 5. Markets reacted instantly: Z.ai shares surged 30% in Hong Kong and are up more than 800% since January.
Contents
GLM-5.2: Technical Specs and Pricing Advantage
GLM-5.2 is an open-weight model that can be freely downloaded, fine-tuned, and run on any infrastructure without developer permission. The system runs entirely on Chinese chips, a critical feature given US export controls.
The cost of using GLM-5.2 is roughly one-sixth that of its US counterparts. DeepSeek V4 Pro charges $3.48 per 1 million output tokens, while Anthropic Fable 5 charges $50 for the same volume. That’s forcing enterprise clients to reassess their relationships with US AI providers.
Z.ai co-founder Tang Jie commented on the Anthropic access suspension:
“This is deeply regrettable. Frontier intelligence should not belong to a few individuals or be subject to sudden rule changes.”
Read more: China and AI — The Sector Consolidates as Anthropic Tightens Data Policy
360 and Sakana AI: ‘China’s Mythos’ and Japan’s Orchestrator
At the ISC.AI 2026 conference in Beijing, 360 founder Zhou Hongyi unveiled two cybersecurity tools under the Yitian Tulong brand. Tulongfeng automates vulnerability discovery in software. The founder explicitly called it “China’s version of Mythos.”
The second product, Yitianzhen, automates cyber defense and incident response. Zhou said vulnerability discovery systems are a strategic state resource and warned of the risk of “one-way transparency,” where American organizations have access to the most advanced models while China does not.
The same week, Japanese startup Sakana AI released its Fugu model. The company said Fugu is “on par with leading models like Anthropic Fable 5 and Mythos Preview.” Unlike the Chinese model, Fugu focuses not only on its own capabilities but also on orchestrating other models through APIs.
Sakana AI co-founder and CEO David Ha said:
“Orchestrator models are the next frontier. Access to the best models can disappear overnight. Collective intelligence is a practical way to hedge against concentration of power.”
Read more: Pew Research — Only 16% of Americans Expect Positive AI Impact
Context: US Export Restrictions and China’s Response
On June 12, the Commerce Department issued a directive restricting exports of Anthropic’s most powerful models, citing national security concerns. Anthropic Fable 5 and Mythos 5 were affected. The trigger was an Amazon request–researchers had found cyberattack instructions in Mythos. On June 26, authorities partially eased restrictions for more than 100 vetted US organizations, but global export rules remain in place.
The restrictions have also pushed Europe to seek its own solutions, but the region lacks computing resources, venture capital, and a developed energy base. The European Chips Act targets 20% of the global semiconductor market by 2030, but independent EU auditors consider that goal unrealistic.
Learn more: What is AI in Crypto? Complete Guide to AI Agents, Tokens & Use Cases
