Amazon Ships Strands Decider 2B, an Open-Source Jev Rival

Amazon has published a roughly 2-billion-parameter decision model built on Alibaba's Qwen3.5-2B base, released free under Apache 2.0 on Hugging Face. VentureBeat reports that Strands Decider 2B hits about 72% accuracy on the JevBench public set (v19), with a 106ms median and 296ms p95 latency on an Nvidia RTX 3090 and roughly 150ms median on an M3 MacBook for small tasks.

The model skips text generation entirely. 'Strands Decider begins with the pretrained Qwen3.5-2B base model. According to AWS, its developers removed the component that predicts the next word and replaced it with a small "pointer" component that scores supplied answer options,' VentureBeat's Carl Franzen writes. A rank-16 LoRA plus the scoring head together add about a million trainable parameters, under an internal architecture AWS calls Hobson.

The target is TypeSafe AI's Jev. 'TypeSafe launched Jev on September 15, calling it a "System One" model' that takes application state and typed questions and returns choices, scores or yes/no probabilities without generating prose. Jev charges $0.042 per million input tokens with no output-token fee, with end-to-end latency reported in the 70 to 500ms range.

'The central idea is simple: if a program needs an answer such as "Should this tool run?" or "Which of these three routes fits this request?", asking a large language model to write an explanation can add time and expense,' the piece explains. Amazon's pitch is that teams can self-host the weights, use the model to route requests, select tools, evaluate outputs or review an agent's actions, and inspect training materials and code.

Franzen does not call it a win. 'AWS's distinguishing proposition is therefore less a demonstrated Jev-killer than an open, reproducible decision layer tied to an existing agent framework,' he writes. A competing open model, Mapika decider-2b v11, still scores 76% on the same benchmark. The article flags that 'adversarial text in an agent's input can influence its verdict' and that AWS has not shown self-hosting beats Jev's hosted price once hardware and ops are counted.

The release lands the same day Cloudflare open-sourced its own Clef decision models, a sign the decision-head genre is maturing into its own layer in the agent stack.