Sakana: Sakana Namazu

Sakana Namazu is a text model from Sakana AI, a provider whose work centers on Japanese language and business contexts. The model is described as a Japanese-specialized reasoning model based on Kimi K2.6, with additional training applied for Japanese language and business use. Its model id on api14 is sakana/sakana-namazu, and it sits in the text category. If you are building for Japanese-speaking users or for workflows where Japanese instruction following matters, this is the model to evaluate first.

On the input side, Sakana Namazu accepts text, images and files, and it produces text output. That combination means you can send a prompt alongside a scanned document, a screenshot or a PDF-style file and get a written response back, rather than being limited to plain text prompts. The context window is 262,144 tokens, which is large enough to hold long documents or extended conversation history in a single request. The maximum output is 65,536 tokens, so responses can be substantial when the task calls for it. Reasoning is a first-class part of the model, and the supported parameters include include_reasoning, reasoning and reasoning_effort, which let you control how much internal reasoning work the model performs before answering.

Those capabilities point to a fairly specific set of use cases. Japanese document review is an obvious fit: send a contract, a report or a form as a file or image, ask questions about it in Japanese, and get an answer in Japanese without a separate translation step. Long-context analysis is another, since the 262,144 token window allows you to work over lengthy materials in one pass. Because the model accepts images as well as files, tasks like extracting information from scanned pages or reading charts and tables inside documents are within reach. Reasoning controls also make it useful for multi-step problems where you want the model to work through the logic rather than answer immediately.

Calling the model through api14 is the same as calling any other model on the gateway, since the API is OpenAI-compatible. You set the model field to sakana/sakana-namazu and send your request to the chat completions endpoint with your api14 key. The supported request parameters are include_reasoning, reasoning, reasoning_effort, structured_outputs, tool_choice, tools and web_search_options. In practice that means you can request structured JSON output through structured_outputs, define functions or external services through tools and tool_choice, tune the depth of reasoning through the reasoning parameters, and enable web search through web_search_options. Because the interface follows the OpenAI schema, existing client libraries and code written for OpenAI models should work with only the model id changed.

Worth keeping in mind is that the output modality is text only. If your application needs generated images or audio, this model will not provide them. The specialization is also a factor: Sakana Namazu is built with Japanese language and business contexts in mind, so for purely English-centric workloads a general purpose model may be a more natural choice. For Japanese tasks that involve long documents, files or images, and where you want control over reasoning depth and structured output, it is a strong candidate.