The Pulse
AWS Releases Open 2B Decision Model for AI Agents
AWS released Strands Decider 2B, an open-source model that selects among predefined answers and returns confidence scores. The model is designed to run locally and handle agent workflow decisions without generating text.
AI.info Team · Written with AI ·
AWS turns routine agent choices into a model
Amazon Web Services released Strands Decider 2B on October 1, an open-source model built to choose among predefined answers rather than generate text. AWS says developers can run it locally and use it in agent workflows where a large language model may be unnecessary for every step.
The model returns a choice and a confidence score. That makes it suited to bounded tasks such as deciding what an agent should do next, while leaving open-ended conversation and more involved reasoning to a general-purpose language model. The approach is meant to give developers a faster, potentially lower-cost decision step without asking a text generator to handle every small judgment.
Marc Brooker built a prototype after seeing Jev
Amazon distinguished engineer Marc Brooker began the project after seeing Jev, a decision model from TypeSafe, and attempting to make a version of his own. His prototype briefly reached the top of the Jevbench ranking for models of its size; AWS engineers then refined it for release through Strands Labs, which works on tools and protocols for deploying AI agents.
Brooker said customer discussions helped expose the need for a narrower tool. “What originally piqued my interest in this class of models was that they make a perfect decider for a workflow step— ‘what is the next thing for me to do here, based on where I am?’” he told TechCrunch. He described the appeal as a workflow step that can be more reliable through confidence scores and a closed set of answers, with lower latency and potentially lower cost.
Choices replace generated text
Strands Decider uses a language-model backbone, but instead of producing a written response, it selects among options supplied for the task. That restriction keeps its answer within the set the developer defines; it does not guarantee that the selected answer is correct. The confidence score gives the surrounding software another signal to consider when deciding whether to continue, ask for review or take another action.
The model is available as an open-source release and is small enough to run locally, according to AWS. The design reflects a division of labor: a decision model handles constrained, repetitive choices, while a broader language model can remain responsible for tasks that need fuller reasoning or generation.
TypeSafe questions whether the field can deliver
Strands Decider arrives as other developers explore the same model category, including TypeSafe’s Jev. The growing number of implementations points to interest in fast, bounded choices for software automation, but it also raises a question about whether strong benchmark results will translate into useful behavior across real tasks.
TypeSafe founder and CEO Diogo Almeida cautioned against treating the new activity as proof of mature competition. “The current batch seems more like ML people wanting to implement a cool architecture than a team deeply dedicated to making intelligence useful,” he told TechCrunch.