Fugu Ultra
Neural Network
Fugu Ultra by Sakana AI is a multi-agent task routing system, Fugu Ultra API for complex scenarios.
Max answer length
(in tokens)
Context size
(in tokens)
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Fugu Ultra?
Fugu Ultra is the highest-performance model in the Fugu family from Sakana AI. It is designed differently from typical monolithic models: it is based on a trained multi-agent orchestration system where a language model routes tasks. Through BotHub, the model is accessible from Russia without a VPN or foreign card, with pay-as-you-go billing in rubles for tokens, no subscription required. With over 250 models in one window, you can compare responses to the same prompt and switch without rewriting your integration: the API is OpenAI-compatible. Chats are encrypted via AES-GCM, and for companies, BotHub provides contracts, invoices, electronic document management, and an admin panel with limits. It is useful when you need to break down a large request into subtasks, understand an unfamiliar codebase, compile an analytical report from disparate materials, automate routine operations via API, or build an agent scenario where Fugu Ultra handles step distribution.Frequently asked questions about Fugu Ultra
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
It is a text model with a reasoning mode, a context window of up to a million tokens, and support for image and document inputs. It is suitable for analyzing large materials, working with and debugging code, analytics, STEM tasks, and agent scenarios with external tool calls.
The output limit is up to 128,000 tokens per response. This is enough for a large code module, a detailed technical document, or a comprehensive analysis, without manually splitting the task into dozens of small requests and stitching pieces together.
Yes, the model's reasoning mode is confirmed. Before the final answer, it breaks down the task step-by-step, which significantly helps in mathematics, logic, data analysis, and code debugging, where the sequence of conclusions is more important than a quick guess.
Yes, the model supports function calling and structured JSON output. This allows for building agents, connecting external services and databases, and obtaining a predictable response format for parsing. In BotHub, everything works via a unified OpenAI-compatible API.
According to our data, the model accepts images and documents as input: you can analyze a diagram, interface screenshot, chart, or multi-page file. Audio and video are not listed in our input formats, so it is easiest to test this scenario in practice.
We do not have precise data on the quality of Russian language performance, so it is most reliable to test the model with your own tasks. In BotHub, this is convenient: access from Russia without a VPN or foreign card, pay-as-you-go billing, and you can switch to another model in one window.