Amazon releases compact model that ranks options with confidence scores

Amazon has released Strands Decider 2B, an open-source decision model inspired by TypeSafe's Jev model. This reflects a growing trend of AI developers creating smaller, specialised models designed for automation tasks rather than general-purpose language generation. The model offers a faster, more cost-effective alternative to large language models for workflows that only need to choose between predetermined options.

Amazon's model is built on Qwen3.5-2B but, unlike traditional LLMs, outputs calibrated decisions with confidence scores rather than generated text. Created by Amazon distinguished engineer Marc Brooker, the model briefly topped the Jevbench ranking for its size class. The release coincided with OpenAI announcing a similar offering, and dozens of researchers have developed comparable decision models since TypeSafe introduced the concept, though TypeSafe's CEO argues that true competition has yet to emerge.