Respan: Span-02

Respan: Span-02 is an open-weight decision model from Respan. It belongs to the decisions category, which means its job is not to write prose or hold a conversation but to answer a question you define about a given state. Respan released it on 16 October 2026, and because the weights are open, you can inspect, self-host or fine-tune the model rather than treating it as a closed black box.

Span-02 accepts text and images as input and produces decisions as output. You supply a state, which can combine written material and pictures, along with the questions you want answered about that state. The model then picks one of several options, or otherwise resolves the question according to the choices you present. The context window is 131,072 tokens, and the maximum output is 117,964 tokens, so both the state you feed in and the decision structure you get back can be large. No request parameters are listed as supported beyond the basics of an OpenAI-compatible call.

The practical uses follow from that shape. Anywhere a system needs a bounded choice rather than free-form text, Span-02 fits: routing an incoming request to one of several queues, classifying a document or screenshot into a category, approving or rejecting a submission against criteria you define, or selecting the next action in a pipeline. Because images are accepted alongside text, the state can include screenshots, scanned forms, diagrams or photographs, which suits review and triage workflows where the evidence is partly visual. The long context lets you pass substantial background material, such as policies, prior records or full documents, without chunking it first.

You call the model through api14's OpenAI-compatible API using the model id "respan/span-02". Point your existing OpenAI client at the api14 base URL, keep your usual chat or completion request structure, and set the model field to that id. The response comes back in the standard format, so no custom parsing layer is needed. Supported parameters are not enumerated for this model, so stick to the common fields and check the live model page if you need something beyond them.

One thing to weigh before adopting it: Span-02 outputs decisions, not explanations or generated text. If your application needs a written answer, a summary or an open-ended conversation, a general text generation model will serve you better. Span-02 is the right pick when the task can be framed as choosing among options you define, and when the input state is text, images or both.