Sakana: Fugu Max

Sakana: Fugu Max is a text generation model from Sakana AI, the lab behind the Fugu family. Fugu Max is described as the cost-performance model in that family, and its architecture is unusual: rather than a single monolithic model, Fugu is a learned multi-agent orchestration system, in which a language model is trained to route work across multiple agents. For a developer, the practical implication is that a single API call to Fugu Max can draw on an orchestrated set of components rather than one fixed model path. It is listed under the text category and was released by the provider on 2026-09-11.

On the input side, Fugu Max accepts text, images and files, and it produces text output. The context window is 1,000,000 tokens, which is large enough to hold very long documents, whole codebases or extended conversation histories in a single request. The maximum output is 128,000 tokens, so long-form generations are possible without chaining many calls. Supported request parameters include include_reasoning, reasoning, reasoning_effort, structured_outputs, tool_choice, tools and web_search_options. Reasoning controls and structured outputs are available alongside tool calling and web search, which means the model can be asked to expose its reasoning, to return responses conforming to a schema, and to invoke external tools or search during a request.

Those capabilities suggest a few concrete use cases. Long-context analysis is a natural fit: feeding large contracts, research papers, logs or repositories and asking for summaries, extraction or question answering within one window. Because images and files are accepted as input, document-heavy workflows such as parsing scanned reports or mixed media material are also reasonable. The tool and web search parameters make it suitable for agent-style applications that need to call functions or look up current information, and structured_outputs helps when the result must be parsed programmatically rather than read by a person. Reasoning effort settings let you trade depth of deliberation against latency for tasks like multi-step planning or debugging.

You call the model through api14's OpenAI-compatible API using the model id "sakana/fugu-max". Because the interface follows the OpenAI chat completions shape, existing clients and SDKs generally work with only the base URL and model id changed. In the request you can set the parameters listed above: include_reasoning, reasoning, reasoning_effort, structured_outputs, tool_choice, tools and web_search_options, along with the usual messages and sampling fields. If you rely on reasoning traces, schema-constrained output, function calling or web search, confirm those parameters are passed explicitly, since behaviour depends on what the endpoint forwards.

The facts given here do not specify latency, throughput or quality comparisons, so it is worth testing Fugu Max against your own workload before committing. If your task is purely text generation with a modest context and no need for tools, search, reasoning controls or structured outputs, a simpler model may be sufficient. Fugu Max makes most sense when you want the large context window, multimodal input or the orchestration and tooling features together in one call.