AWS debuts Strands Decider 2B, a first lightweight decision model for accelerate agentic workflows

Amazon Web Services (AWS) has unveiled Strands Decider 2B, a lightweight decision model aimed at speeding up agentic AI workflows. The open-source decision model eliminates the need for text generation, reducing token consumption and response latency. AWS designed Strands Decider 2B for local deployments, allowing developers to run it on their laptops and in public cloud environments.

By providing a dedicated decision-making engine, AWS aims to accelerate the pace of AI development. Decision models, or "System 1 models," make choices based on predefined options without generating text. They assign confidence scores to each decision, indicating its accuracy. AWS's Strands Decider 2B is built on top of a standard LLM (Qwen3.5-2B) but replaces the traditional LLM head with a compact pointer head containing just 1 million parameters.

The model has been fine-tuned using a rank-16 LoRA adapter to score hidden states directly against answer positions. Strands Decider 2B has shown strong performance in benchmarks, matching the accuracy and calibration of other open-source 2B models on JevBench. The model is available for download on Hugging Face, with full codebase, training scripts, and examples on GitHub.

AWS hopes the AI developer community will explore Strands Decider 2B to enhance agentic tasks like model routing, tool selection, context management, guardrail enforcement, and policy classification, potentially leading to the creation of hybrid agents that combine decision models and LLMs.