Amazon launches Strands Decider 2B, a fast open-source decision model

Amazon announced on Tuesday that it has released Strands Decider 2B, a free, fast, open‑source decision model that rivals the recently popular Jev. The new model can pick from a list of options in less than a second and ships with downloadable weights that developers can run on anything from a laptop to a server farm. The announcement came on the same day Amazon’s AI blog highlighted the growing demand for lightweight agents that can make quick choices without generating long text.

The push for decision‑only models follows a shift in AI design. Generative models like GPT‑4 are powerful but costly and slow, especially when they need to answer simple yes‑no or multiple‑choice questions. TypeSafe AI’s Jev, released two weeks ago, showed that a model trained to choose from a fixed set of options can be cheaper and faster. Strands Decider 2B builds on that idea, offering a more efficient architecture that reduces inference time and GPU memory usage by up to 50 % compared with Jev.

Technical details show the model uses a lightweight transformer with 30 million parameters and a custom pruning scheme that keeps the core decision logic intact. Benchmarks on an NVIDIA A100 show latency of 12 ms per query and a throughput of 200 decisions per second, all while consuming less than 2 GB of GPU RAM. The open‑source release invites researchers to tweak the weights and add new decision trees, and early adopters report that the model can be integrated into AWS Lambda functions with minimal overhead.

Looking ahead, Amazon plans to embed Strands Decider 2B into its SageMaker and Bedrock services, making it available to developers who want to build cost‑effective AI assistants, recommendation engines, or automated workflows. The open‑source nature of the model could spur a new wave of community‑driven decision agents, lowering the barrier to entry for small businesses and accelerating the adoption of AI in everyday applications.