Amazon Web Services’ Strand Labs has launched Strands Decider 2B, a new open-source decision model inspired by TypeSafe’s Jev, aimed at improving automation in AI-driven workflows. Unlike large language models that generate text, this model focuses on efficiently selecting from predefined options and providing confidence scores for its choices. Strands Decider 2B is lightweight enough to run locally, offering users a speedier and more cost-effective alternative for decision-making tasks within automated systems.

The idea for Strands Decider 2B originated from Amazon engineer Marc Brooker, who developed a prototype after reviewing Jev, eventually leading to a polished release by Amazon’s Strand Labs. Brooker highlights that many AWS customers do not require the complexity or expense of large-scale language models for certain workflow decisions. Instead, Strands Decider 2B fits well as a specialized component that reliably determines next steps in processes, balancing accuracy, calibration, and speed.

Built on the Qwen3.5-2B LLM framework, the model employs a “torso” of a large language model to deliver calibrated decisions instead of generating text outputs. The name Jev pays homage to economist William Stanley Jevons, reflecting the belief that lowering the costs and barriers to intelligence can expand demand and use cases. The surge of similar models following TypeSafe’s announcement demonstrates broad interest, though it also raises questions about their differentiation and practical value.

Despite the influx of Jev-like models, TypeSafe’s CEO Diogo Almeida believes the real challenge remains in achieving genuinely intelligent and useful decision-making systems. He cautions that current efforts by others often focus on architectural replication rather than deeply enhancing intelligence. Meanwhile, Brooker emphasizes fine-tuning models like Strands Decider 2B to maintain both rapid decision-making and linguistic understanding as crucial for their success, especially in niche markets where development costs are far more accessible than for frontier AI labs.