[Digital Today reporter Chi-gyu Hwang (황치규)] Amazon Web Services (AWS) has open-sourced Strands Decider 2B, a lightweight decision-making model used for agentic AI development.

SiliconANGLE reported on Wednesday that the model is designed to reduce token consumption and response delays by operating without generating text and selecting one option from a predefined set.

It is optimized for local deployment and rapid experimentation. Developers can run it on a laptop or in a public cloud.

AWS' strategy is to support the development of "hybrid agents" in which a dedicated decision engine lets the decision model handle selections while an LLM handles complex reasoning.

Decision-making models have a different structure from large language models (LLMs). While LLMs generate text, code, images and video, decision models choose one of several predefined options. Each decision comes with a confidence score, allowing users to gauge response accuracy. But it cannot explain its reasoning because it does not generate text.

AWS sees recently emerged Type-Safe AI Jev as showing the potential of decision models, but believes performance limits exist in complex reasoning due to its parallel output structure. Strands Decider 2B is an attempt to address those limits, SiliconANGLE said.

The base model is Qwen3.5-2B. AWS applied a custom "pointer head" of about 1 million parameters to Qwen3.5-2B instead of the existing LLM head.

The version released this time is v.20, with 2 billion parameters. AWS explained that 2 billion parameters is small enough to run on local hardware with less than 150 milliseconds of latency while still capable of handling complex decisions.

The model can be downloaded from Hugging Face. The full codebase, training scripts and examples are provided on GitHub.