Asian AI Startups Launch Mythos-like Models Amid US Export Bans

Asian AI startups are developing frontier models to bypass US export restrictions

Asian AI companies are launching high-capability models to provide alternatives for businesses and governments unable to access US-based frontier AI. This trend is driven by the US government's recent ban on the export of Anthropic's Mythos and Fable 5 models to non-Americans, creating a strategic vacuum in the Asian market.

Sakana AI launches Fugu for orchestration and sovereignty

Tokyo-based Sakana AI has released Fugu, a frontier AI model designed specifically for agentic workflows and the orchestration of other models via APIs. Sakana AI positions Fugu as a way for Japanese businesses and government agencies to reduce their reliance on US technology and avoid the risks associated with tightening export controls.

Key characteristics of Fugu include:

Orchestration Focus: CEO David Ha describes Fugu as an "Orchestration Model," arguing that the ability to coordinate multiple models is the next frontier beyond simply increasing model size.
Strategic Hedging: The company views Fugu as a practical hedge against the concentration of AI power in a single provider, noting that access to top-tier models can disappear overnight due to political mandates.
Local Optimization: The model is optimized for Japanese language and culture, utilizing an approach that works effectively with smaller datasets.

China's 360 develops cybersecurity-specific AI

Chinese cybersecurity firm 360 has unveiled Tulongfeng, an AI tool designed to compete directly with Anthropic's Mythos. Unlike Sakana's hedging strategy, 360's approach is focused on national strategic assets.

360 has released two primary security tools:

Tulongfeng: Specialized in the automatic discovery of software vulnerabilities.
Yitianzhen: Designed to automate cyber defense and incident response.

Founder Zhou Hongyi has characterized vulnerability-finding AI as a national strategic asset, warning against "one-way transparency," where only certain nations possess advanced vulnerability-detection capabilities.

Market Reaction and Technical Skepticism

While the companies claim their models are "Mythos-like" in capability, the technical community has expressed significant skepticism due to a lack of transparent, third-party benchmarking.

Performance and Cost Concerns

Users reporting early experiences with Fugu have noted issues with cost and efficiency. Some users claim that the model is significantly slower and more expensive than Anthropic's Opus, with some reporting that a $20 plan was exhausted in a single prompt for complex tasks.

Architectural Questions

There is ongoing debate regarding whether Fugu is a monolithic model or a routing system. Some analysis suggests Fugu Ultra may be a "learned multi-agent orchestration system" that routes tasks to a pool of underlying models rather than being a single large language model.

The "Mythos-like" Ambiguity

Critics argue that the term "Mythos-like" is functionally meaningless without public benchmarks on platforms like the LMSYS Chatbot Arena. As one observer noted:

"Without reliable benchmarks, they are Mythos-like only in the sense that they accept text as input and produce text as output."

The Broader Impact of Export Controls

The US ban on Mythos and Fable 5 has accelerated the drive toward "AI sovereignty" in Asia. While Sakana AI's leadership maintains that US models remain important and that the current shift is not necessarily a permanent realignment, the emergence of local alternatives trained on local nuances suggests a growing gap in the global AI ecosystem. This shift is further complicated by the high run-rate revenue of US labs—Anthropic reported a run-rate of $47 billion in May 2026—which may be at risk if Asian enterprise customers pivot to local providers.