Amazon releases open-source decision model Strands Decider 2B

A small model for decision steps

Amazon Web Services (AWS) has introduced Strands Decider 2B, an open-source model aimed at the decision steps inside AI agent workflows. Rather than generating new text like classic large language models, it selects among predefined options and returns a confidence score for its choice.

Inspired by Jev, released under Strands Labs

The project started with Jev, a model built by TypeSafe. After seeing Jev, AWS engineer Marc Brooker began an experimental effort that briefly topped the Jevbench ranking in its size category. Amazon engineers then developed the project further and published it under Strands Labs, the company's umbrella for agent tools and protocols.

Why it exists

According to Amazon, the need also surfaced in conversations with AWS customers. Some steps in agent workflows don't require the broad capabilities of large language models, yet using big models there can mean higher latency and cost. Strands Decider 2B is positioned for narrower tasks, such as an agent choosing its next action.

Built on Qwen3.5-2B

The model is based on Qwen3.5-2B, but instead of generating text it is calibrated to make decisions among specific options. Fully open source, it is small enough to run on local devices.

A market splitting in two

Dozens of similar models have appeared since Jev's debut, and OpenAI announced its own Decisions API this week. TypeSafe CEO and founder Diogo Almeida says that despite the wave of models, the company doesn't see serious competition yet, arguing the real challenge is making such models genuinely useful and smart. The trend points to a split: large models keep handling planning, analysis and content generation, while small decision models take on routing, classification and next-action selection.