Amazon Web Services has released Strands Decider 2B, an open-source decision model inspired by TypeSafe’s Jev model. Announced on October 1, 2026, the model enables developers seeking more fit-for-purpose intelligence for computer automation to quickly and cost-effectively choose among predefined options and measure decision confidence. The fully open-source model is small enough to be ready to use and run locally. It was released in the same week as OpenAI’s announcement of a similar offering.
The project grew out of a personal effort by Amazon distinguished engineer Marc Brooker after he saw Jev. The effort briefly rose to the top of the Jevbench ranking among models of its size. Amazon engineers later refined the model and released it through Strands Labs, which develops new AI tools and protocols. According to Brooker, AWS customers’ agentic workflows did not always need the capabilities or cost of a full-featured large language model. With its closed response space and confidence scores, Strands Decider can provide a more reliable, lower-latency, and potentially cheaper workflow step.
The model is built on the “torso” of a Qwen3.5-2B-based large language model that makes calibrated choices instead of generating text. TypeSafe took the name Jev from economist William Stanley Jevons; the name alludes to the idea that demand could increase as the cost of computer intelligence falls. TypeSafe CEO and founder Diogo Almeida said that developing intelligent models remains difficult despite the proliferation of similar models, and that he does not currently see any real competition. The company is focusing on developing its future models. Brooker, meanwhile, said that the key challenge is preserving language understanding and general knowledge capabilities while improving accuracy and calibration.
Why it matters
This development shows that, alongside the approach of turning to systems that generate comprehensive text for every task in AI use, narrower models are also being positioned for tasks with predefined boundaries. This distinction could affect the choices developers building agentic workflows make among cost, latency, and controllability; the ability to run the model locally also diversifies the conditions of use. The open-source license lowers the access barrier for teams that want to examine decision models and adapt them to new tools. However, how accurate the model’s confidence scores will remain across different tasks, and to what extent its selection-focused structure will preserve language understanding and general knowledge capabilities, remain open questions.
Background
Amazon is not a new name in the FikirPilot archive: we have published 11 reports mentioning the name in the past 90 days; the latest was dated 26 September 2026.
Term: large language model
A large language model is software trained on vast amounts of text that generates the continuation of the text it is given based on probability; it does not write what it knows, but the most likely continuation.