Amazon releases its own Jev clone as decision models flood the web

Amazon Web Services' research arm Strands Labs has released Strands Decider 2B, an open-source decision model designed to choose among pre-defined options and emit calibrated confidence scores. Built on the Qwen3.5-2B ‘‘torso’’ rather than being optimized to generate freeform text, the model prioritizes speed, low cost, and reliability for closed-domain decision steps and is small enough to run locally.

The project began as a homegrown effort by AWS distinguished engineer Marc Brooker, who started experimenting after seeing TypeSafe’s Jev. Brooker’s prototype performed well enough on Jevbench for its size that AWS engineers formalized and cleaned it up into the Strands Labs release. According to Brooker, customer conversations revealed a demand for decision-focused models in agentic workflows where a full-featured LLM is unnecessary or too expensive.

Strands Decider 2B arrived the same week OpenAI announced a comparable offering, underscoring a wider flurry of activity around Jev-style models. Proponents argue these decision models deliver lower latency and cost while providing useful confidence measures; critics and some founders caution that many recent efforts are experimental implementations rather than deeply engineered attempts to make models broadly useful and smart.

AWS and TypeSafe participants alike say the key technical trade-off will be preserving accuracy and calibration for fast decision making without eroding the broader capabilities that make a model generally useful—such as multilingual understanding and background knowledge. TypeSafe’s leadership says they continue to iterate on their models and view many copycat efforts as not yet rivaling their work.