Respan: Span-01 Lite

Respan: Span-01 Lite is a behavior scoring model from Respan, released on 26 September 2026. It is the free, lighter tier of Span-01, the provider's larger behavior scoring model. Rather than a general purpose chat model, Span-01 Lite belongs to the decisions category: its job is to evaluate text against behaviors you define and return a judgment about them. If you are building a pipeline that needs a consistent, repeatable read on user generated or model generated text, this is the kind of model you would reach for.

The model takes plain language text as input and returns decisions as output. You define each behavior in plain language, and for every behavior the model returns the probability that the behavior is present in the text you supplied. That is the core contract: you describe what you are looking for in words, and the model scores how likely it is that the text exhibits it. No context length is published for this model, and no supported request parameters are listed, so treat it as a fixed, opinionated scoring endpoint rather than something you tune per call. The maximum output is listed as zero, which reflects that the model does not produce free form prose; its output is the decision itself.

The practical uses follow directly from that shape. Moderation and policy checks are an obvious fit: define behaviors such as harassment, spam or self promotion, and score incoming text against them. Evaluation of model outputs is another, where you define behaviors like factual hedging, refusal or instruction following and use the probabilities to compare generations or catch regressions. Product analytics and research work benefit too, since you can turn a qualitative question into a numeric signal across a large corpus without writing a classifier for each new behavior. Anywhere you would otherwise train a bespoke classifier, defining the behavior in plain language and letting Span-01 Lite score it is a lighter path.

You call it through api14's OpenAI-compatible API using the model id "respan/span-01-lite". Because the interface follows the OpenAI schema, existing SDKs and clients work with a change of base URL and model id. The model supports no additional request parameters beyond the basics of the call, and it accepts text input only, so keep prompts to plain text and expect decisions rather than generated prose in the response.

Since Span-01 Lite is the lighter tier, it is the sensible default when you want behavior probabilities at low friction. If you need a broader or more capable scoring model, the full Span-01 tier from the same provider is the natural next step. If your task is generation or conversation rather than scoring, a text output model will fit better, because Span-01 Lite returns decisions, not text.