Japan Just Entered the AI Race with Sakana, Claiming to Beat Mythos with a Router

On June 12, 2026, the U.S. government blocked global access to Anthropic's Claude Mythos 5. Tokyo-based startup Sakana AI responded by releasing Fugu. Fugu matches the performance of Mythos, but it does so without relying on a single massive model. It uses a router to coordinate smaller, specialized AI agents.

How Fugu routes tasks

Fugu is a multi-agent orchestration system. When a user sends a prompt, Fugu evaluates the request. It then assigns the task to the most capable model available in its active pool.

This collective intelligence approach produces fast, high-quality responses. By stitching together the outputs of several specialized models, the system competes directly with restricted frontier models.

Figure 1: Fugu uses a lightweight selection head to dynamically route tasks to a pool of various underlying AI models.

Bypassing export controls

The June restrictions exposed a vulnerability for companies relying on U.S.-based frontier models. A single regulatory decision can remove access overnight.

Fugu provides vendor independence. Because the system can swap out the underlying models in its pool, it maintains performance even if individual models face new export controls. This creates a sovereign-friendly solution for enterprises and nations seeking to avoid dependency on a single AI provider.

Figure 2: The architecture emphasizes collective intelligence over brute-force scale.

References

• Sakana AI — Official documentation on the Fugu orchestration system.
  sakana.ai
• VentureBeat — Coverage of Sakana AI's launch and the response to US export controls.
  venturebeat.com