Amazon Releases Open-Source Strands Decider 2B Decision Model for Local Deployment

Amazon's Strands Agents team announced on October 1st the release of its Strands Decider 2B decision-making model. The model is now available as open-source on GitHub, with its weights accessible on Hugging Face. Notably, Strands Decider 2B is designed to run locally on common consumer hardware, including both CPUs and GPUs.

The Strands Decider 2B model is built upon a pre-trained Qwen3.5-2B "backbone." It features a specialized "head" designed for scoring, which replaces the original language model's text generation capabilities. This new head is significantly smaller, with just over one million parameters, while the backbone has been fine-tuned using a rank-16 LoRA adapter.

In performance evaluations using the JevBench dataset, Strands Decider 2B demonstrated strong accuracy and calibration. It ranked third among models in its 2B parameter class, outperforming other competitors strictly within that size limit.

When run locally on standard hardware, such as an NVIDIA GeForce RTX 3090 graphics card, the model achieved a median decision latency of 113 milliseconds for small decision tasks.