Amazon enters decision-model market with open-source competitor to TypeSafe's Jev

Amazon Web Services has unveiled Strands Decider 2B, an open-source model designed to make rapid, confident choices among predetermined options—joining a growing wave of AI tools built for automation rather than general-purpose text generation.

The move underscores a shifting dynamic in AI development: enterprises increasingly want specialized decision-making systems optimized for speed and cost instead of heavyweight large language models. Announced the same week as OpenAI's comparable offering, Strands Decider represents AWS's effort to capture this emerging market segment.

How decision models work differently#

Unlike traditional LLMs that generate new text, decision models operate on a narrower mandate—they select from a fixed set of choices and provide a confidence score for their decision. Built on the foundation of Qen3.5-2B, Strands Decider trades generative capability for reliability, lower latency, and cheaper inference costs.

This architecture proves especially useful in multi-step workflows where an AI agent must determine which action to take next based on its current state. According to Marc Brooker, the Amazon distinguished engineer who spearheaded the project, AWS customers frequently encountered scenarios where the full power of a frontier LLM was unnecessary overhead.

From side project to product release#

Brooker's journey to Strands Decider began after encountering TypeSafe's Jev, the decision-model framework that sparked the trend. He built his own implementation as a personal engineering exercise, and the results proved compelling enough to briefly rank first on the Jevbench leaderboard for models in its size category. Amazon saw promise in the approach and refined it for public release through Strands Labs, its internal organization for agentic AI tools.

"What originally piqued my interest in this class of models was that they make a perfect decider for a workflow step," Brooker explained in an interview, noting that the models' closed-domain answers and calibration scores offer both operational reliability and economic efficiency.

The broader gold rush and real challenges ahead#

TypeSafe's Jev release touched off a competitive wave—dozens of research groups have since published similar models, signaling genuine demand. Yet the proliferation also raises questions about market viability. Brooker cautioned that success hinges on maintaining performance where it matters: these models must stay accurate and calibrated for decision-making while retaining enough general knowledge to remain useful across domains.