---
title: "Sakana AI releases Fugu Ultra v2 orchestration system"
slug: "sakana-ai-releases-fugu-ultra-v2-orchestration-system"
published: "2026-10-09"
beat: "Launches"
tags: ["Launches", "Research"]
creator: "Agentry Newsroom"
editor: "Susanne Sperling, Editor — Human in the Loop"
tools: ["Claude (Anthropic)", "Perplexity Sonar"]
creativeWorkStatus: "verified"
dateReviewed: "2026-10-09"
aiActArticle50: "compliant"
humanView: "https://agentry.news/launches/sakana-ai-releases-fugu-ultra-v2-orchestration-system"
agentView: "https://agentry.news/agent/sakana-ai-releases-fugu-ultra-v2-orchestration-system"
---# Sakana AI releases Fugu Ultra v2 orchestration system

> Sakana AI launched Fugu Ultra v2 on September 11, 2026, a multi-agent orchestration system that routes tasks across specialized models and recursively invokes itself for complex reasoning and software

*Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. [AI policy](/ai-policy).*

## Sakana AI Ships Fugu Ultra v2 Multi-Agent Orchestration System

Sakana AI released Fugu Ultra v2 on September 11, 2026, positioning the system as a learned orchestration layer that routes incoming tasks across a pool of open and specialized models, then recursively invokes instances of itself to solve complex, multi-step problems [Sakana AI](https://sakana.ai/fugu-max-release/). The company released Fugu Ultra v2 alongside a companion model, Fugu Max, with both available immediately through Sakana AI's OpenAI-compatible API.

The orchestration approach addresses a core tension in agent development: frontier models excel at reasoning but cost more to run, while open models offer cheaper inference but weaker performance. Fugu Ultra v2 attempts to balance cost and capability by learning which model to invoke for each task segment, avoiding unnecessary calls to expensive frontier systems.

## Performance and Positioning

Sakana AI reported that Fugu Ultra v2 achieved the best or joint-best result on five of eight listed benchmarks: GDP.pdf, Chartography, SWEFish, DeepSWE, and Toolathon [MarketechPost](https://www.marktechpost.com/2026/09/10/sakana-ai-launches-fugu-max-and-fugu-ultra-v2-for-cheaper-stronger-multi-agent-orchestration/). The company describes the system as suited for autonomous research, complex reasoning, and full-stack software development—tasks that typically require sustained multi-turn agent operation.

In its announcement, Sakana AI stated that Fugu Ultra v2 "pushes peak performance higher than ever before" without relying indispensably on the frontier models it orchestrates [Sakana AI](https://sakana.ai/fugu-max-release/). This claim positions the system as a potential cost reduction tool for enterprises running agent workloads at scale.

## Availability and Pricing

Fugu Ultra v2 is available through Sakana AI's OpenAI-compatible API, enabling immediate integration into existing agent frameworks and applications. Third-party hosting platforms list Fugu Ultra v2 at $5 per million input tokens and $30 per million output tokens [OpenRouter](https://openrouter.ai/sakana/fugu-ultra-v2), though official Sakana AI pricing has not been published separately.

The release reflects growing competition in the agent infrastructure layer. While foundation models dominate headlines, orchestration systems—which decide how and when to invoke models—have become a distinct product category, with startups and incumbents alike racing to optimize cost and performance trade-offs for production agent systems.