Sakana AI is inviting early testers for Sakana Fugu, its flagship commercial AI product. This multi-agent orchestration system aims to achieve state-of-the-art performance by coordinating pools of frontier foundation models for tasks like coding, mathematics, and scientific reasoning.
Sakana Fugu, initially available as an API, has served as an internal tool for Sakana AI's researchers and engineers. The company is now opening applications for external beta testers. You can apply for the beta test on their website.
Pushing Boundaries with Collective Intelligence
Sakana AI's core philosophy centers on the belief that the most capable AI systems will emerge from collections of specialized agents working collaboratively, rather than monolithic models scaled in isolation. This principle underpins their previous research, including evolutionary model merging, the AI Scientist, ShinkaEvolve, and AB-MCTS.
Sakana Fugu represents the productized form of this research direction, offering a solution to the economic inefficiencies and optimization difficulties inherent in managing multiple specialized foundation models.
Traditional approaches often require users to juggle numerous API keys for different models, each excelling in distinct areas. Sakana Fugu resolves this by dynamically orchestrating a diverse pool of powerful models. Instead of relying on pre-defined domain knowledge for team organization, Fugu learns to dynamically assemble agents and coordinate them through efficient, often non-obvious, collaboration patterns. This adaptive orchestration, based on Sakana AI's ICLR 2026 papers 'Trinity' and 'Conductor,' yields superior performance on established benchmarks.
The daily brief
Get the free daily brief with the day's AI news, new AI startups and funding rounds every morning.
Join 64,000 other subscribers.
The system demonstrates significant gains over individual models in tasks such as GPQAD, LCBv6, and SWEPro, as shown in performance tables comparing Fugu Mini and Fugu Ultra against leading models like Gemini, GPT, and Opus.
Using Sakana Fugu
Sakana Fugu is accessible via APIs, featuring compatibility with standard OpenAI-format endpoints. This allows for straightforward integration into existing workflows for users already employing models like GPT, Gemini, or Claude. Behind the familiar interface, Sakana Fugu automates the coordination across its model pool, handling collaboration topology, role assignment, and subtask dispatch.
Two variants are offered: Sakana Fugu Mini, optimized for low latency, and Sakana Fugu Ultra, designed for maximum performance on demanding tasks. This Multi-Agent Orchestration System offers a flexible approach to leveraging diverse AI capabilities.
Join the Beta
Sakana AI is seeking researchers and engineers across various domains to participate in the beta test. The goal is to assess Fugu's performance in untested areas, identify limitations, and gather insights into user needs. Involvement is encouraged for those using foundation model APIs in coding assistants or seeking performance advantages in their engineering and business projects.
