Sakana AI Unleashes Fugu Max & Ultra v2: Revolutionizing Multi-Agent AI Orchestration for Cost & Capability

Optimizing AI: The Dual Challenge of Capability and Cost

For readers tracking the shift, In the rapidly evolving world of artificial intelligence, deploying powerful models often presents a dual challenge: achieving top-tier performance while simultaneously managing operational costs. Many organizations find themselves overspending on complex, multi-trillion-parameter models for simple tasks, or compromising on capability to stay within budget. Sakana AI is stepping forward with an innovative solution to this dilemma, introducing two new models in its Fugu family: Fugu Max and Fugu Ultra v2.

Meanwhile, These aren’t traditional foundation models; instead, they are sophisticated, learned orchestrators designed to intelligently route AI workloads across a diverse pool of other models through a single, unified API. The goal? To dynamically select the most cost-effective machinery that can still successfully complete a given task, pushing the boundaries of what’s possible in multi-agent AI.

Understanding the Fugu Orchestration Approach

At its core, Sakana Fugu operates as an intelligent traffic controller for AI tasks. When a query is presented, the Fugu model, described as a language model in its own right, dynamically constructs an ‘agentic scaffold’. This scaffold then assigns roles and delegates sub-tasks to the most appropriate underlying models from its extensive pool.

In practical terms, Sakana AI frames this challenge as a two-axis problem, visualized through the Pareto frontier. This economic concept illustrates the trade-off: enhancing quality typically increases cost, while reducing cost often diminishes quality. Fugu Max and Fugu Ultra v2 leverage the same core orchestration architecture, but their optimization targets diverge, allowing businesses to choose based on their primary need.

The Technology Underpinning Fugu

The Fugu system is built on cutting-edge research, drawing insights from ICLR 2026 papers. Key components include:

• TRINITY: A lightweight, evolved coordinator that assigns specific roles—Thinker, Worker, or Verifier—across different turns of a task.
• The Conductor: Trained with reinforcement learning, this component discovers natural-language coordination strategies and generates focused prompts, ensuring efficient task execution.

For example, The training methodology combines large-scale fine-tuning, evolutionary algorithms, and reinforcement learning, creating a highly adaptive and efficient system.

Fugu Max: Maximizing Output Per Dollar

Fugu Max is engineered for cost-efficiency, aiming to deliver the best possible output for every dollar spent. It achieves this by significantly expanding the pool of models it can orchestrate, incorporating a wide array of open-weight and specialized models, including the NVIDIA Nemotron family through a collaboration with NVIDIA. Fugu Max intelligently routes each task to the leanest model capable of solving it.

Key Highlights of Fugu Max:

• Pricing: Highly competitive at $2 per 1 million input tokens and $6 per 1 million output tokens.
• Cost Efficiency: Sakana AI reports output prices 40% to 60% lower than models like Sonnet 5, GPT 5.6 Terra, and Kimi K3.
• Performance: Achieved the best overall score on 6 benchmarks, including Terminal Bench 2.1, GPQA Diamond, AA-LCR, GDP.pdf, AutomationBench, and SWEFish.
• Efficiency Frontier: Expanded the cost-performance Pareto frontier on 7 out of 10 benchmarks.

That said, This positions Fugu Max as a compelling option for organizations seeking elite model performance at a substantially lower cost, often 2x to 6x less than current market leaders.

Fugu Ultra v2: Raising the Ceiling for Complex Reasoning

For workloads demanding the highest capabilities, Fugu Ultra v2 steps in. This model is meticulously designed for complex reasoning, autonomous research, and full-stack software development. It demonstrates its most significant advancements in sustained reasoning, particularly when dealing with visual and structured data.

Key Highlights of Fugu Ultra v2:

• Chartography: Scored 48.3 in visual reasoning and data interpretation, significantly outperforming Opus 5 (27.3) and Fable 5 (29.5).
• DeepSWE: Achieved 74.3 in real-world software engineering tasks, surpassing models priced 3x to 5x higher per token.
• Breadth & Consistency: Ranked best or joint-best on 5 of 8 benchmarks (GDP.pdf, Chartography, SWEFish, DeepSWE, Toolathon) and within the top two on 7 of 8 benchmarks.

Interestingly, Notably, Fugu Ultra v2 achieves these impressive results without relying on proprietary models like Fable 5, Fable 5.1, or GPT-6-Astra in its agent pool. This strategic independence reduces exposure to vendor lock-in, API revocations, and sudden service cutoffs, offering greater reliability and flexibility.

Deployment and Availability

Both Fugu Max and Fugu Ultra v2 are available today as hosted APIs, accessible through Sakana’s OpenAI-compatible API. Existing users can integrate them with a simple one-line switch. Notably Sakana AI does not offer open weights for self-hosting, and the service is not currently available in the EU/EEA region.

Conclusion: A New Era for AI Workload Optimization

However, Sakana AI’s Fugu Max and Fugu Ultra v2 represent a significant leap forward in multi-agent AI orchestration. By intelligently balancing capability and cost, these models empower businesses to deploy AI solutions more efficiently and effectively than ever before. Whether the priority is maximizing output per dollar with Fugu Max or tackling the most complex reasoning tasks with Fugu Ultra v2, Sakana AI offers a flexible and powerful platform to navigate the complexities of modern AI deployment.

Expert Perspective

A practical read on Sakana AI Fugu Orchestration starts with fugu. That is where the earliest effects are likely to show up if this development keeps building.

What happens next will come down to adoption speed, policy response, and execution quality. That combination could make Sakana AI Fugu Orchestration a meaningful reference point across models.

For decision-makers, the useful lens is not the headline alone but how ultra changes priorities once organizations have to respond.

Frequently Asked Questions

Why is Sakana AI Fugu Orchestration important?

Optimizing AI: The Dual Challenge of Capability and CostFor readers tracking the shift, In the rapidly evolving world of artificial intelligence, deploying powerful models often presents a dual challenge: achieving top-tier performance while simultaneously managing operational costs.

What impact could Sakana AI Fugu Orchestration have?

Many organizations find themselves overspending on complex, multi-trillion-parameter models for simple tasks, or compromising on capability to stay within budget.

What should readers watch next with Sakana AI Fugu Orchestration?

Sakana AI is stepping forward with an innovative solution to this dilemma, introducing two new models in its Fugu family: Fugu Max and Fugu Ultra v2.Meanwhile, These aren’t traditional foundation models; instead, they are sophisticated, learned orchestrators designed to intelligently route AI workloads across a diverse pool of other models through a single, unified API.

How does this relate to fugu?

It connects because the article frames fugu as one of the clearest areas where the topic may be felt in practice.

Source: https://www.marktechpost.com/2026/09/10/sakana-ai-launches-fugu-max-and-fugu-ultra-v2-for-cheaper-stronger-multi-agent-orchestration/