Amazon Releases Open-Source Decision Model Strands Decider 2B

Amazon has released Strands Decider 2B, an open-source decision model designed to make decisions in fractions of a second. This model is part of Amazon's experimental agent-development project, Strands Labs.

The central idea behind this model is that instead of asking a large language model to write an explanation for every decision, a decision model can score allowed answers directly and quickly. This makes it useful as a frequent, narrow checkpoint around a more capable agent.

Strands Decider 2B begins with the pretrained Qwen3.5-2B base model, which was adapted by removing the component that predicts the next word and replacing it with a small 'pointer' component that scores supplied answer options. The architecture of this design is called 'Hobson'. Unlike a chatbot, it makes one forward pass and returns a distribution over the available choices.

Amazon says its team iterated rapidly and will publish version 20 alongside the code, training data, and scripts. Developers can download Strands Decider 2B from Hugging Face and use it under an enterprise-friendly Apache 2.0 license. They must supply their own hardware or cloud computing resources to run it.

AWS says its model can make local decisions in tens of milliseconds on short tasks and under 100 milliseconds on some common hardware and inputs. The accuracy and calibration of the probabilities using the public portion of JevBench were measured by AWS, with a chart tracing successive versions through version 19, which improved over the team's earlier checkpoints.

Strands Decider ranks second among public models around its size on this test, and first among those publishing a complete training recipe. Amazon is offering only the open-source model for now, with no hosted API or per-call charge.