Amazon releases Strands Decider 2B, an open-source take on TypeSafe's Jev

Amazon Web Services released Strands Decider 2B on Thursday, an open-source decision model inspired by TypeSafe's Jev. Instead of generating text, it sorts between pre-decided options and reports how confident it is in its choice. OpenAI previewed a similar offering, its Decisions API, the same week.

The model grew out of a project by Marc Brooker, a distinguished engineer at Amazon, who built his own take on Jev. His homebrew version briefly topped the JevBench ranking for models of its size, and Amazon released a cleaned-up version through Strands Labs. Brooker said the need emerged in talks with AWS customers.

How it works

Strands Decider 2B builds on Alibaba's open Qwen3.5-2B model. AWS removed the text-generating head and replaced it with a pointer head that scores the supplied answer options; that head has just over a million parameters, and the backbone uses a rank-16 LoRA adapter. Restricting the answer space keeps it from inventing options nobody offered.

In AWS's example, a user asks for the weather without naming a city. The agent guesses a location and proposes calling a weather tool; before the call runs, Decider checks whether the arguments are grounded in the conversation. The check runs through Strands' intervention system, where developers can proceed or deny the call. In the demo, Decider runs locally.

Results

On JevBench's public set, AWS says Strands Decider ranks second among public models of roughly 2 billion parameters, and first among public models with a full training recipe available. It answers every easy-tier question correctly, and decisions take under 100 milliseconds on an Nvidia RTX 3090. The release includes the training data and scripts.

Why it matters

Jev kicked off a wave of similar models a few weeks ago, followed by projects such as Kev, imajev and Laya, with big vendors joining in. Brooker says the challenge is optimizing fast decision-making without compromising intelligence. With smaller markets, building something interesting costs hundreds or thousands of dollars.

TypeSafe is keeping its head down on 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 Diogo Almeida told TechCrunch. Whether AWS will offer a hosted version remains open.

Sources: TechCrunch · The New Stack · Techmeme