RespanAI's open-weight decision model Span-02 launches on OpenRouter

RespanAI has launched Span-02, its open-weight decision model, on OpenRouter. The model accepts text and image inputs, allowing users to define their questions and receive outputs such as a choice, a yes/no/unknown answer, or a score with probabilities.

Input tokens are priced at $0.01 per million, while output is free. The model supports a 131K context window. According to RespanAI, Span-02 is the cheapest decision model available; with thinking off, it delivers frontier-level decision-making at low latency. Users can also assign a reasoning budget, allowing the model to reason until the deadline and return its best decision so far, with the option to stop anytime.

Span-02 comes with fully open weights and a Rust inference engine that RespanAI says is 7 times faster than vLLM. An auto mode uses an adaptive learned gate to select the reasoning budget for each decision.