Fugu: Single API, Multi-Agent Power Rivals Anthropic

Sakana Fugu: The Multi-Agent Orchestration System That Rivals Anthropic’s Fable and Mythos

Sakana AI has officially launched Fugu, a groundbreaking multi-agent orchestration system that functions like a single model while internally coordinating a dynamic pool of frontier large language models. Unlike traditional AI solutions that rely on a single vendor, Fugu intelligently routes tasks across models from providers such as Claude, GPT, and Gemini, verifies their outputs, and synthesizes a unified answer behind one OpenAI-compatible API.

How Fugu Operates as a Seamless Single Model

From the user's perspective, Fugu behaves exactly like a standard language model with a single endpoint for all requests. Internally, however, it acts as a sophisticated router and coordinator that decides when to solve a task directly and when to assemble a team of expert agents. This system eliminates the complexity of building multi-agent workflows from scratch, as Fugu learns to coordinate delegation, communication, and verification without hard-coded roles.

Core Capabilities of the Orchestrator

• Dynamic Delegation: Fugu selects the most appropriate model for each specific step of a task.
• Recursive Operations: The system can call upon instances of itself recursively to handle complex reasoning chains.
• Output Verification: Every result is checked against others before being combined into a final answer.
• Unified Interface: Users interact with a single API while enjoying the aggregated power of multiple top-tier models.

Two Variants for Different Performance Needs

Sakana Fugu is available in two distinct variants, both accessible through the same OpenAI-compatible API to suit different use cases and performance requirements.

Fugu: Optimized for Speed and Everyday Tasks

The standard Fugu variant balances strong performance with low latency, making it ideal for everyday coding assistance, code review, chatbots, and tools like Codex. A key feature of this version is its flexibility, which allows teams to exclude specific agents from its pool to meet strict data privacy, security, or compliance requirements.

Fugu Ultra: Tuned for Maximum Quality on Complex Problems

Fugu Ultra is designed for the most demanding, multi-step challenges where maximum answer quality is critical. This variant coordinates a deeper and fixed pool of expert agents, ensuring consistent performance for hard problems in engineering, science, and reasoning. Unlike the standard version, the agent pool in Fugu Ultra is fixed, meaning opt-out options are not available.

Benchmark Performance Matching Anthropic’s Top Models

Sakana AI states that its Fugu models stand shoulder-to-shoulder with Anthropic’s Fable 5 and Mythos Preview on key benchmarks. Fugu Ultra specifically matches the performance of these top-tier Anthropic models in difficult areas such as engineering, scientific reasoning, and logical deduction. This achievement demonstrates that a learned multi-agent orchestrator can outperform individual frontier models without the need to train another massive foundation model.

Advantages Over Traditional Single-Vendor Models

The primary advantage of Fugu lies in its adaptability and independence from single-vendor dependencies. The system can interchange agents as needed, ensuring continued functionality even if a specific model encounters export restrictions or access challenges. This flexibility provides a strategic edge for organizations that require reliable AI capabilities without the risk of being locked into a single provider’s ecosystem.

Conclusion

Sakana Fugu represents a major shift in AI architecture by proving that a learned orchestrator can effectively manage a pool of diverse models to deliver frontier-level capabilities. By offering both a speed-optimized version for daily tasks and a quality-optimized version for complex reasoning, Fugu provides a versatile solution that rivals the performance of Anthropic’s leading models while offering greater flexibility and vendor independence.

For developers and enterprises seeking powerful AI without single-vendor dependency, Fugu offers a robust, scalable, and intelligent alternative that simplifies the complexity of multi-agent systems into a single, easy-to-use interface.