Rayline releases lightweight AI model ARC for task routing

On November 12, 2026, TestingCatalog (@testingcatalog) reported that Rayline has launched ARC, a lightweight AI model designed to dynamically choose which model should handle each step of a task and how deeply that step should be reasoned about. According to the report, users can send a task to ARC and have it automatically assign easy steps to a lighter model and harder ones to a stronger model, so that overall execution can be made faster.

A second TestingCatalog post from the same day added more detail on how people can actually use it. Users can either have ARC run the chosen model and send back the answer, or pick the best model for the task themselves and run it on their own setup. That second option points to a more self-hosted, do-it-yourself workflow, where ARC is used as a dispatcher for tasks of varying complexity rather than as the model that produces every answer.

Both reports come from the same account and the same time on November 12, 2026, so the available public record so far consists only of these two posts. There is no further information in the material about ARC's technical specs, benchmarks, licensing, or availability beyond what is described above.