AWS open-sources Strands Decider 2B, joining the rush of small Jev-style decision models

TechCrunch reports that Amazon Web Services released Strands Decider 2B from its Strands Labs group. Built on a Qwen3.5-2B base, which the report calls the model's "torso," it outputs calibrated choices among a fixed set of options, with a confidence score for each. It is open source and small enough to run locally, and it is aimed at steps in agent workflows that require a decision but not a full frontier language model.

AWS distinguished engineer Marc Brooker started the project after seeing Jev, the decision model from startup TypeSafe. An early homebrew version briefly topped Jevbench, the benchmark for this class, in its size category. The release joins the team's earlier Strands Harness agent framework.

Decision models have multiplied quickly. The same week, OpenAI announced its own version, which coverage framed as a way to keep its swarming agents in check. Nokia open-sourced AnyJev, a training-free layer for turning open models into calibrated deciders, and Contrastive-LM released CLM-8B, which it says scores agent actions faster than Jev.

TypeSafe named Jev after economist William Stanley Jevons, reflecting a thesis that cheaper intelligence increases demand for decisions inside automated systems. CEO Diogo Almeida told TechCrunch he still does not see real competition, arguing many clones are interesting architectures rather than dedicated work on decision quality.

For agent developers, a free local decider is a practical tool for closed-domain branching: routing a ticket, choosing a tool or deciding whether to escalate, with a confidence score that can trigger human review when low. Running such steps on a small model can cut cost and latency compared with calling a frontier model for every choice.

The value depends on calibration. A confidence score that does not reflect actual accuracy is worse than none, because it encourages automation where it is not warranted. Comparative results against Jev on public benchmarks, and production case studies where confidence scores change routing, will show whether the category matures. More in our agents coverage and tools coverage.