Reasoning Model
A model architecture that performs intermediate reasoning steps before producing an output, rather than direct pattern completion.
A reasoning model works through intermediate steps, context, constraints, and partial conclusions, before it produces a final output, instead of directly completing a prompt in one pass the way a standard generation model does. Those intermediate steps are what let it check something against a rule or a source partway through, rather than only after the fact.
SwishX's production model, Finsen, is built this way. A request moves through staged reasoning, decoding the brief, harnessing brand context, linking claims, routing to a generation model, verifying the output, before anything actually renders, rather than turning a prompt directly into a finished video. That staged structure is also what a model-routed request sits inside: routing to a specific downstream generation model is one stage in a longer reasoning sequence, not the whole of it.