Overview
- Strands Decider 2B, which AWS released Thursday, October 1, 2026, is an open‑source decision model built to pick from developer‑supplied options and emit calibrated confidence scores.
- The model uses a Qwen3.5‑2B 'torso' with a lightweight pointer head that scores predefined answers instead of generating free‑form text, a design that reduces token use and speeds responses.
- AWS published the full training recipe, data and code so teams can run and inspect the model locally, and it placed second among public ~2B models on JevBench while topping the list of models that include a full recipe.
- AWS reported sub‑100 millisecond decisions on an NVIDIA RTX 3090 and about 150 ms median on an M3 MacBook for small tasks, highlighting tradeoffs between speed, calibration, and real‑world robustness.
- The release follows TypeSafe’s Jev and a wave of similar 'decision' models, and it raises practical questions for developers about whether to run deciders locally or use hosted APIs and how to balance cost, latency, and reliability.