AWS Releases Open-Source Strands Decider 2B Decision Model for Agentic Workflows

The model is available for download on Hugging Face, with its full codebase, training scripts and examples on GitHub, according to SiliconANGLE. AWS optimized it for local deployments and rapid experimentation, so developers can run it on laptops as well as in public cloud environments. Decision models, sometimes called System 1 models, differ from large language models: instead of producing text, code, images or video, they select from predefined choices and assign a confidence score to each decision.

SiliconANGLE reported that Strands Decider 2B is built on top of a standard LLM using the Qwen3.5-2B base torso. AWS replaced the usual LLM head with a customized pointer head of about 1 million parameters and fine-tuned the torso with a rank-16 LoRA adapter to score hidden states of available choices directly against answer positions. TechCrunch reported the base as Qen3.5-2B. AWS iterated on the model repeatedly; the released version is named v.20. The company chose the 2-billion-parameter scale because it can run locally with less than 150 milliseconds of latency while still making complex decisions, SiliconANGLE reported.

AWS said Strands Decider 2B achieved high accuracy and calibration on JevBench compared with other open-source 2B models, according to SiliconANGLE. TechCrunch reported that the homebrew project behind the release briefly reached the top spot on the JevBench ranking for models of its size. The model is AWS's first real attempt to improve on decision models, SiliconANGLE reported.

The release follows the emergence of TypeSafe AI Inc. and its Jev model. Jev was designed to make fast, structured decisions that software and AI agents can use directly. AWS's team believes Jev has structural shortcomings, including performance issues when asked to perform intricate reasoning because of its parallel output structure, SiliconANGLE reported. Amazon distinguished engineer Marc Brooker came up with the Strands project after seeing Jev and building his own take on the model, according to TechCrunch.

Brooker told TechCrunch that customer conversations showed agentic workflows did not always require the capability or cost of a fully featured LLM. 'What originally piqued my interest in this class of models was that they make a perfect decider for a workflow step—what is the next thing for me to do here, based on where I am?' he said. He said the model offers a workflow step that can be structured more reliably because of confidence scores and the closed domain of answers, with lower latency and potentially lower cost. Brooker also said AWS must balance accuracy and calibration against language understanding and general knowledge. He said he did not expect frontier labs to dominate the space, particularly because smaller markets mean the cost to build something interesting is in the hundreds or thousands of dollars.

TypeSafe CEO and founder Diogo Almeida told TechCrunch that the company was keeping its head down and improving future models. 'I get that people think it is a gold rush, but they might be underestimating the difficulty of making the models actually smart,' he said. Almeida said he did not yet see real competition for TypeSafe, adding that the current batch of models seemed more like machine learning practitioners wanting to implement a cool architecture than a team deeply dedicated to making intelligence useful. TechCrunch reported that OpenAI announced a similar offering the same week and that dozens of similar models have been produced by researchers since TypeSafe debuted its idea.

AWS said it hopes the AI developer community will experiment with Strands Decider 2B and accelerate agentic tasks including model routing, tool selection, context management, guardrail enforcement and policy classification. The company said the model also paves the way for hybrid agents that use decision models for simple, repetitive choices and LLMs for complex reasoning.