Amazon Web Services has released a free, open source AI decision model called Strands Decider 2B. The model was inspired by Jev, a new type of AI model from the startup TypeSafe.
The release came the same week OpenAI announced its own similar model. AI developers have been looking for tools that are better suited to computer automation than large frontier language models.
How Strands Decider Works
Strands Decider 2B is built to sort between options that have already been set. It then gives a score that shows how confident it is in its choice.
The model is available now. It is small enough for users to run on their own machines.
Like other decision models, it is built on the "torso" of a large language model. In this case, that base is Qwen3.5-2B. Instead of writing text, it delivers calibrated choices.
TypeSafe named its model Jev after the economist William Stanley Jevons. His theory says that when something becomes cheaper, such as computer intelligence, demand for it can actually rise.
From Side Project to Amazon Release
Amazon distinguished engineer Marc Brooker started the project after seeing Jev. He then tried to build his own version of the model.
His homemade version did well enough to briefly reach the top spot on the Jevbench ranking for models of its size. Amazon engineers then cleaned it up and released it through Strands Labs, a group that builds tools for deploying AI agents.
Brooker said the need came up in talks with AWS customers. Their automated workflows did not always need the power or cost of a full language model.
"What originally piqued my interest in this class of models was that they make a perfect decider for a workflow step," Brooker told TechCrunch. He described that step as deciding "what is the next thing for me to do here, based on where I am?"
He said the model can make a workflow step more reliable because of its confidence scores and its limited set of answers. He added that it can also be faster and possibly cheaper.
Researchers have built dozens of similar models since TypeSafe introduced its idea. Brooker said the challenge is making the model fast at decisions without hurting its intelligence.
"There is a very careful balance to be found," he said. He explained that developers want to improve accuracy and calibration without weakening the model's language skills and general knowledge.
Brooker said he does not expect frontier labs to dominate this area. He noted that building a useful model for smaller markets can cost only hundreds or thousands of dollars.
TypeSafe executives say they are focused on improving future models. CEO and founder Diogo Almeida told TechCrunch that people "might be underestimating the difficulty of making the models actually smart."
Almeida said he does not yet see real competition for his company. "The current batch seems more like ML people wanting to implement a cool architecture than a team deeply dedicated to making intelligence useful," he said.
