Amazon Web Services releases an open source decision-making model inspired by TypeSafe’s Jev, and AI developers are increasingly seeking intelligence better suited for computer automation than Frontier LLM.
Amazon’s Strands Decider 2B, released the same week that OpenAI announced a similar product, is a fast, low-cost way to categorize predetermined options and provide a measure of how confident you are in your choice. This model is completely open source, available today, and small enough to run locally.
Marc Brooker, a prominent engineer at Amazon, came up with this project after looking at Jev and trying to build his own take on such a model. The homebrew project was successful enough to briefly take the top spot in the Jevbench rankings for similarly sized models, so Amazon engineers cleaned it up and released it as a product from Strands Labs, an organization that develops new tools and protocols for deploying AI agents.
According to Brooker, the need for such a tool emerged in conversations with AWS customers whose agent workflows didn’t always require the functionality or cost of a full-featured LLM.
“I was originally interested in this class of models because they perfectly determine steps in a workflow, ‘Based on where I am, what should I do here next?'” Brooker told TechCrunch. He said this provides customers with “workflow steps that can be constructed in a more reliable way because of the confidence score, because of the closed field of answers, (and) with lower latency and potentially lower costs.”
Like other decision-making models, Strands Decider is built on the “body” of the LLM (Qen3.5-2B in this case), but instead of generating text, it provides tailored choices. TypeSafe named the model Jev after economist William Stanley Jevons. This hopes to support his theory that the falling cost of things like computer intelligence may actually increase demand for it.
The fact that dozens of similar models have been created by researchers since the TypeSafe idea was announced shows widespread interest, but also raises questions about how valuable they are. Brooker suggests that the challenge will be to optimize the model’s rapid decision-making without compromising the model’s intelligence.
“It requires a very careful balance to improve accuracy and coordination performance on these kinds of tasks without reducing performance on understanding and knowledge of different languages. That’s what makes it versatile, interesting and useful,” he told TechCrunch.
Still, I don’t necessarily expect Frontier Labs to dominate this space, especially in smaller markets where it costs hundreds or thousands of dollars to build something interesting.
Meanwhile, TypeSafe executives say they are keeping their heads down and improving future models.
“I understand that people think it’s a gold rush, but they may be underestimating the difficulty of actually making the models smart,” CEO and founder Diogo Almeida told TechCrunch, adding that so far no real competition has emerged for his company.
“The current team looks more like ML people who want to implement a good architecture than a team that is passionate about making intelligence useful.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
