Asian AI Startups Build Resilient Models Amid US Export Limits

On June 22, 2026, Tokyo-based Sakana AI introduced Fugu, a new AI orchestration system. It came just ten days after the US government imposed export controls that pulled Anthropic’s Mythos and Fable 5 models offline.

Fugu is named after the Japanese blowfish. But unlike a single AI model, it works by orchestrating multiple AI agents. It calls on a pool of frontier models through their APIs. This lets it split tasks among specialists and merge their outputs into one answer.

Sakana AI calls Fugu “not a monolithic model like Claude or GPT. It is an orchestration model.” This means it does not rely on just one AI system. Instead, it routes requests dynamically. If a model supplier restricts access or goes offline, Fugu bypasses the issue.

Fugu comes in two versions: standard Fugu and Fugu Ultra. Fugu Ultra is designed to compete with Anthropic’s best, including Mythos Preview and Fable 5. Sakana’s tests show Fugu Ultra scored 73.7 on SWE-Bench Pro and 95.5 on GPQA-D, plus 93.2 on LiveCodeBench.

These benchmark results suggest Fugu Ultra performs on par with top models. But it does not always beat every individual model in its pool. This pool includes major commercial players like Google Gemini 3.1 Pro, OpenAI GPT-5.5, and Anthropic Claude Opus 4.8.

Why Build an Orchestration Model?

David Ha, CEO of Sakana AI, explained the motivation behind Fugu. He said relying on a single AI provider for national infrastructure is risky. The export controls on Anthropic models made this risk impossible to ignore.

Fugu aims to solve this by providing resilience. It can dynamically route around restrictions or outages. This means no single provider’s disappearance can take the system down. It also means users can keep accessing frontier AI even when some models are unavailable.

Ren Ito, Sakana AI’s co-founder, added, “AI should not become a technology that is hoarded; it should be one that is developed together.” This philosophy drives Fugu’s design to integrate multiple models and share capabilities.

China’s 360 Joins the Race

On the same day Sakana announced Fugu, Chinese cybersecurity firm 360 unveiled Tulongfeng. This AI tool claims it can compete directly with Anthropic’s Mythos. While details remain limited, 360’s move signals strong interest in building alternative AI models in Asia.

Both Sakana and 360’s efforts reflect a broader push in Asia to develop advanced AI amid US export restrictions. These new models seek to challenge US dominance and reduce dependence on single suppliers.

Fugu is closed-source and partly relies on closed APIs. Sakana has not disclosed the full pool of models behind it. It can cost up to $10 per message for complex tasks. The system is not available in the European Union or European Economic Area yet.

Despite these limits, early users see promise. Patrick, director at Yarra Web, said about their AI system Viktor, “Viktor is now an integral team member, and after weeks of use we still feel we haven’t uncovered the full potential.”

Fugu’s multi-agent approach may signal a new way to build AI systems. Instead of one giant model, orchestration could become the key to resilience and performance. As export controls squeeze access to certain US models, tools like Fugu could redefine AI’s future across Asia and beyond.