📰 Key Takeaways

With the US government ban ongoing, two Asian AI companies have launched alternatives that can benchmark against Anthropic’s top models. Chinese cybersecurity firm 360 unveiled the “Dragon Slayer Phoenix,” claiming it can go head-to-head with Anthropic’s Mythos. In the same week, Tokyo AI startup Sakana AI released the Fugu model, named after the pufferfish, with official claims that its capabilities are “on par with top-tier models like Fable 5 and Mythos Preview.” Fugu’s core design is positioned for Agent collaboration—it can coordinate and调度 other models via API. Founder David Ha pointed out that “orchestration models are the next frontier beyond just scaling up model size.”

Sakana was co-founded in 2023 by former Google researchers Ren Ito, Llion Jones, and David Ha. They’re known for building generative AI that runs on small datasets and is optimized for Japanese language and culture. While the company stresses that Fugu’s launch timing was “purely coincidental,” their website already clearly markets “frontier capabilities with no export control risk,” targeting Japanese enterprises and government agencies looking to reduce their exposure to export controls. That said, Sakana acknowledged that “American models are still crucial for the Asian market” and doesn’t view this launch as a permanent shift away from US AI. Instead, they frame the current situation as a temporary structural adjustment opportunity.

💬 JudyAI Lab’s View

With US export controls tightening, Asian AI vendors are rapidly filling the market gap—360’s Dragon Slayer Phoenix and Sakana AI’s Fugu are both benchmarking against Anthropic’s top models. This is a concrete signal that geopolitics is reshaping the AI supply chain.

What deserves our attention isn’t the capability claims themselves, but Fugu’s design philosophy: centered on “orchestration models,” where AI coordinates and schedules other models rather than relying on a single large model to do it all. Sakana founder David Ha clearly stated that orchestration is the next frontier beyond model scaling—which directly relates to how we think about multi-Agent system architectures. Even more noteworthy is Sakana’s market positioning: championing “no export control risk,” precisely targeting Japanese enterprises and government agencies looking to reduce policy exposure. They’re quickly securing a position within a structural opportunity window, not making a long-term claim to fully replace US models.

If you’re planning multi-Agent workflows, consider reevaluating your “orchestration layer”—a collaborative architecture that can coordinate and schedule other models might be worth prioritizing over simply chasing bigger models.

📅 Original Source Info

🔗 Further Reading