ScaleDown: Classify-1

ScaleDown: Classify-1 is a structured decision model from ScaleDown, served through the provider's Decisions API. Rather than generating free-form prose, it is built to answer questions you define in advance, which places it in the decisions category rather than the general chat category. If you are evaluating models on api14, the practical question is not whether it writes well but whether your task can be framed as a choice, a yes/no question, or an ordered scale.

The model accepts text, images, and files as input, so you can send a passage of text, a scanned page, or a document and have it evaluated directly. Its output modality is decisions: it scores the supplied input against your caller-defined questions and returns calibrated probabilities. The context window is 1,000,000 tokens, and the maximum output is 900,000, which means long documents and large batches of material can be handled in a single request without chunking. There is no free-text generation to parse, so you get structured answers rather than sentences you have to interpret.

That shape of output points to a fairly specific set of uses. Routing and triage are natural fits: deciding which queue a ticket belongs to, or whether an incoming message meets a yes/no condition before it reaches a human. Document review is another, since files and images are accepted, so classification of forms, invoices, or scanned records against a fixed set of categories is possible. Ordered-scale questions suit sentiment or severity grading, and because the model returns probabilities rather than a single label, you can set your own confidence threshold and send low-confidence cases elsewhere. Any workflow where you need a consistent, repeatable judgement across many items is a better match than one where you need an explanation written out.

You call it through api14's OpenAI-compatible API using the model id "scaledown/classify-1". Because the interface is OpenAI-compatible, existing clients and SDKs work with a change of base URL and model id. Note that the supported parameters list is empty, so you should not assume that sampling controls such as temperature or top_p are accepted; send the model id, your input, and the questions you want scored, and rely on the defaults. Check the live price table on this page for current rates.

The main limitation to weigh is the output format itself. Classify-1 returns decisions, not prose, so if your application needs a written summary, a conversational reply, or generated content of any kind, a general text model will fit better. It is also worth confirming that your questions can be expressed as choices, yes/no, or an ordered scale, since that is the framing the model is designed around. Within those bounds, and with a context window this large, it is a reasonable choice for high-volume classification where you want probabilities you can act on rather than text you have to parse.