AWS Releases Strands Decider 2B, an Open-Source Decision Model for AI Agents

Unlike general-purpose Large Language Models (LLMs) that generate arbitrary text, decision models—sometimes called "system one" models—focus on specific tasks such as picking between options or assigning numerical scores. This specialization allows the model to operate with significantly lower latency and cost. The Strands Decider 2B is built on the Qwen3.5-2B architecture but replaces the traditional language modeling head with a "pointer head" that scores the provided options.

AWS distinguished engineer Marc Brooker, who led the project, noted that the model addresses a need among customers whose agentic workflows do not always require the high cost or reasoning capabilities of a full-scale LLM. The model is capable of running locally on standard hardware, with median latencies reported around 115ms on an Nvidia RTX 3090 and 153ms on an M3 MacBook.