Amazon open-sources Strands Decider 2B for agentic workflows

The model gives AWS a local, lower-cost option for choosing between fixed actions instead of generating text.

Amazon Web Services has open-sourced Strands Decider 2B, a compact decision model inspired by TypeSafe’s Jev that is optimized for choosing among predefined options rather than generating text. Built on the Qen3.5-2B LLM torso, the model returns calibrated choices with confidence scores and is small enough to run locally, offering a low-latency, lower-cost component for agentic workflows. AWS distinguished engineer Marc Brooker started the project after experimenting with Jev, and internal work on it led to a polished release from Amazon’s Strands Labs. Brooker says customer demand came from workflows that don’t need the expense or capability of full frontier LLMs at every step, and he frames the main technical challenge as improving accuracy and calibration without sacrificing general language understanding. TypeSafe CEO Diogo Almeida, whose company introduced Jev and the Jevbench ranking, argues that many recent decision models look like experiments in a trendy architecture rather than serious attempts to make these systems truly intelligent, and he does not yet view them as strong competition.