Fugu Blends Models Task By Task

2026-07-03 · Source: The Batch | DeepLearning.AI | AI News & Insights - www.deeplearning.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, medium

Summary

Tokyo-based research lab Sakana AI released Fugu and Fugu-Ultra, two new models designed to orchestrate other models and agents via a single API. Fugu handles discrete tasks like basic coding and chat, while Fugu-Ultra specializes in long-running tasks such as extensive coding and research. These systems deliver performance comparable to Claude Mythos 5 and GPT-5.6 Sol, notably without relying on a single underlying model. Fugu and Fugu-Ultra outperformed Claude Fable 5, Claude Mythos Preview, and GPT-5.5 on benchmarks including Terminal Bench 2.1, GPQA-Diamond, and LiveCodeBench. Fugu-Ultra further achieved top results on SWE-Bench Pro, Humanity's Last Exam, and LiveCodeBench Pro. The models accept text and images as input, producing text output, and are available through the Sakana API, OpenRouter, and Vercel. Pricing for Fugu is based on the underlying model's rate, while Fugu-Ultra has token-based and subscription plans.

Key takeaway

For AI Scientists and ML Engineers seeking to reduce vendor dependency and optimize performance across varied tasks, Fugu and Fugu-Ultra present a compelling alternative to single-model solutions. You can utilize these orchestration models to dynamically select the best underlying LLM for each subtask, potentially achieving top-tier results while maintaining control over data handling and cost. Evaluate their API for your specific discrete or long-running agentic workflows.

Key insights

Model orchestration, by delegating tasks to specialized agents, achieves superior performance and reduces single-vendor reliance.

Principles

• Orchestrating diverse models can surpass individual model capabilities.
• Task-specific model selection enhances overall system efficacy.
• Agentic systems can operate at higher levels of abstraction.

Method

Fine-tune an LLM to orchestrate agentic workers by scoring worker model outputs, then train it with an evolutionary algorithm (sep-CMA-ES) for model selection. For complex tasks, use a coordinator model (Conductor) trained via GRPO to break down and manage subtasks.

In practice

• Employ Fugu for discrete tasks, Fugu-Ultra for complex, long-running ones.
• Configure model pools to manage data sensitivity and optimize costs.

Topics

• Model Orchestration
• Agentic AI
• Fugu Models
• LLM Benchmarking
• AI System Design
• Vendor Independence

Best for: AI Engineer, CTO, VP of Engineering/Data, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Sakana AI's Fugu orchestrates multiple LLMs to match Anthropic's Fable and Mythos benchmarks — Fugu orchestrates multiple LLMs to match top benchmarks and reduce single-provider dependence.
• Issue 360 — AI advancements emphasize model orchestration, specialized training, and robust safety measures amidst increasing government oversight.
• TAI #207: Claude Opus 4.8 Is Better, but Dynamic Workflows Are the Bigger Story — Dynamic Workflows, where LLMs orchestrate multi-agent tasks, represent a significant shift towards scalable, complex problem-solving, despite increased token costs.
• Japan’s New AI Just Challenged Claude Mythos — But There’s a Catch — Coordinating multiple specialized AI models can achieve high performance but introduces complexity and potential for suboptimal outcomes.
• The Agentic Economy: From Lone Agents to Orchestrated Swarms — High-reliability AI requires specialized, orchestrated agent swarms, not single general-purpose models.
• [AINews] not much happened today — Frontier model constraints drive multi-model orchestration and advanced agent infrastructure for robust AI applications.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Batch | DeepLearning.AI | AI News & Insights - www.deeplearning.ai.