AWS open-sources Strands Decider 2B, a small decision model inspired by TypeSafe's Jev

Amazon Web Services has open-sourced a small decision model called Strands Decider 2B, the latest entry in a wave of releases inspired by TypeSafe's Jev. According to TechCrunch, the model is available now, small enough to run locally, and built to sort between pre-decided options while reporting how confident it is in its choice — a faster, cheaper alternative to routing every step of an automated workflow through a full-size LLM.

The release arrived in the same week that OpenAI announced a similar offering, a sign of how quickly the category has moved from research idea to shipping product.

From side project to official release

The project began as a personal experiment. Amazon distinguished engineer Marc Brooker told TechCrunch he built his own take on the decision-model concept after seeing Jev, and the homebrew version performed well enough to briefly hold the top spot on Jevbench, a ranking for models of its size. Amazon engineers then cleaned it up and released it through Strands Labs, a group within the company developing new tools and protocols for deploying AI agents.

Brooker said the idea came out of conversations with AWS customers, whose agentic workflows did not always need the capability or the cost of a fully featured LLM at every step. These models, he argued, make "a perfect decider for a workflow step" — answering what to do next based on where the workflow currently is, with reliability that comes from confidence scores and a closed set of possible answers, at lower latency and potentially lower cost.

What a decision model actually is

Strands Decider is built on the "torso" of an existing LLM — a 2-billion-parameter model that TechCrunch identifies as Qen3.5-2B — but instead of generating text it emits calibrated choices. TypeSafe named its original model after the economist William Stanley Jevons, whose theory holds that when the cost of something falls, such as computer intelligence, demand for it can rise.

Since TypeSafe debuted the idea, researchers have produced dozens of similar models. TechCrunch reads that spread as evidence of broad interest in the approach — and of an open question about how much value each new entrant actually adds.

The tuning trade-off

Brooker's central technical concern is balance. Optimizing hard for accuracy and calibration on decision tasks risks degrading the underlying model's language understanding and general knowledge, which are what make it broadly useful in the first place. Striking that balance, he suggested, is the key engineering challenge for the category.

He also doubts that the frontier labs will come to dominate it. The markets here are smaller, and building something interesting costs hundreds or thousands of dollars — low enough for individuals and small teams to compete.

TypeSafe is unfazed

TypeSafe, for its part, says it is keeping its head down and improving future models. "I get that people think it's a gold rush, but they might be underestimating the difficulty of making the models actually smart," CEO and founder Diogo Almeida told TechCrunch, adding that he does not yet see real competition for his company. The current batch of clones, in his words, seems "more like ML people wanting to implement a cool architecture than a team deeply dedicated to making intelligence useful."

Why it matters

The release reflects a shift in how AI gets deployed. Agentic workflows need many small, dependable decisions rather than one generative powerhouse at every step, and a 2B-parameter model that outputs choices with confidence scores fits that job — and that budget — better than a frontier LLM. Open-sourcing it, and making it small enough to run locally, means those decision points can be run and inspected by anyone, and its brief run atop the Jevbench ranking suggests it is competitive at its size. The unresolved question, acknowledged on both sides, is whether the flood of Jev-style models will translate into genuinely better automation or a long tail of architectural copycats. AWS putting its weight behind one signals that at least the major cloud providers see real customer demand for the pattern.