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

Overview of the Shift Toward AI Sovereignty

Asian AI startups are launching high-capability models to fill the vacuum created by US government export controls. Following a US ban preventing non-Americans from accessing Anthropic's Mythos and Fable 5, companies in Japan and China have introduced tools designed to match these capabilities, framing their releases as a strategic hedge against the risk of sudden access loss to US-based frontier AI.

Sakana AI's Fugu: An Orchestration-First Approach

Sakana AI, a Tokyo-based startup founded by former Google researchers, has released Fugu, a model designed to provide frontier-level capabilities without the risk of export controls. Rather than relying on a single monolithic model, Fugu is positioned as an "orchestration model" designed to coordinate agent usage across multiple other models via APIs.

Key Characteristics of Fugu

• Agentic Design: Fugu is built to orchestrate access to other models, acting as a coordinator for AI agents.
• Regional Optimization: The model is optimized for the Japanese language and culture, utilizing smaller datasets to maintain affordability.
• Strategic Positioning: Sakana AI describes Fugu as a "practical hedge" against the concentration of AI power in a few US providers, arguing that national infrastructure should not rely on a single provider whose access can "disappear overnight."

China's 360: Cybersecurity-Focused AI

Chinese cybersecurity firm 360 has unveiled Tulongfeng, an AI tool specifically designed to automatically discover software vulnerabilities. Alongside Tulongfeng, the firm also released Yitianzhen, a tool for automating cyber defense and incident response.

Strategic Implications for China

360's founder, Zhou Hongyi, has characterized vulnerability-finding AI as a "national strategic asset." He warned against "one-way transparency," where only certain global actors possess advanced vulnerability-detection capabilities, making the development of local alternatives to Mythos a matter of national security.

Community Analysis and Technical Skepticism

While the companies claim their models are "Mythos-like," the technical community has expressed significant skepticism regarding these claims due to a lack of transparent, third-party benchmarks.

Performance and Cost Concerns

Users reporting early experiences with Fugu have highlighted issues with performance and cost:

"I tried the Fugu models... the result was worse than Opus, incredibly slow, and I ended up exhausting the new 5 hour window... it hardly created something opus was able to do at a fraction of the time and cost."

Architectural Skepticism

Some analysts suggest that Fugu may not be a standalone frontier model but rather a routing system. According to reports from OpenRouter, Fugu Ultra is described as a "learned multi-agent orchestration system" that routes tasks across a pool of underlying models. This has led to theories that users are paying a markup for API calls to other models, explaining the high cost and slow response times.

The "Mythos-like" Benchmark Gap

Critics argue that without public benchmarks on platforms like the LMSYS Chatbot Arena, the term "Mythos-like" is a marketing term rather than a technical specification. Because Mythos is currently unavailable to many, the claim is difficult to disprove, leading to calls for a UN-sponsored standards body to handle global AI benchmarking.