Amazon Web Services (AWS) has introduced an open-source decision model, taking inspiration from TypeSafe’s Jev, as AI developers increasingly seek intelligent solutions better suited for computer automation than extensive frontier LLMs.
Amazon’s Strands Decider 2B, unveiled in the same week OpenAI announced a similar product, offers a rapid and economical method for selecting among predetermined options and provides a confidence score for its choice. This model is entirely open-sourced, readily available, and compact enough to operate on local systems.
Marc Brooker, an Amazon distinguished engineer, initiated the project after encountering Jev and attempting to develop his own version of such a model. His personal project achieved significant success—briefly holding the top position on the Jevbench ranking for models of its size. Consequently, Amazon engineers refined it and released it through their Strands Labs, an entity dedicated to creating novel tools and protocols for deploying AI agents.
Brooker explained that the necessity for such a tool arose from discussions with AWS clients, whose agentic workflows didn’t consistently demand the full capabilities or expense of a comprehensive LLM.
“My initial interest in this category of models stemmed from their ideal fit as a decider for a workflow step—‘what is my next action here, given my current state?’” Brooker shared with TechCrunch. He emphasized that it provides customers “a workflow step that can be structured for enhanced reliability, thanks to the confidence scores and limited answer domain, while also being lower latency and potentially more cost-effective.”
Similar to other decision models, Strands Decider is built upon the foundational structure, or “torso,” of an LLM—specifically Qwen3.5-2B. However, instead of generating text, it delivers finely tuned choices. TypeSafe named their model Jev in homage to economist William Stanley Jevons, hoping to evoke his theory that a reduction in the cost of a resource—like computational intelligence—can paradoxically increase its demand.
The fact that dozens of comparable models have been developed by researchers since TypeSafe first introduced its concept highlights widespread interest, but also raises questions about their ultimate value. Brooker suggests that the primary challenge will involve optimizing the model’s rapid decision-making without compromising its overall intelligence.
“There’s a critical equilibrium to strike where you aim to boost its accuracy and calibration for these tasks, without diminishing its ability to comprehend various languages or its existing knowledge base, which is what makes it versatile and valuable,” he informed TechCrunch.
Nevertheless, he doesn’t anticipate frontier labs will necessarily dominate this sector, especially given that for smaller markets, the expense of developing something compelling can be in the hundreds or thousands of dollars.
For their part, TypeSafe executives state they are focusing on continuous improvement for future models.
“I understand people perceive it as a gold rush, but they might be underestimating the complexity of making models truly intelligent,” CEO and founder Diogo Almeida told TechCrunch, asserting that for now, he doesn’t observe genuine competition for his company emerging yet.
“The current offerings seem more like machine learning enthusiasts eager to implement an innovative architecture rather than teams profoundly committed to making intelligence genuinely useful.”