Article
CommunityJev Models & Reimplementations
Strands Decider 2B — strandsagents.com
The Strands Agents team's 2B open-source decision model: it swaps Qwen3.5-2B's language-model head for a roughly million-parameter pointer head that scores offered options, runs on local CPU or GPU, and placed third of 33 in its JevBench class.
- Format
- Article
- Reading time
- 2 min
- Published by
- Community
- Author
- Marc Brooker, Mike Chambers, Fabio Nonato de Paula
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Strands Decider 2B — huggingface.co.
Used inJev for Software & DevTools
- Capabilities
- Open sourceRuns locally
Highlights
- Takes Qwen3.5-2B, removes the LM head and adds a ~1M-parameter pointer head that scores each offered option.
- Median decision latency around 115ms on an RTX 3090; small tasks about 153ms on an M3 MacBook.
- Reported 3rd of 33 in the 2B class on JevBench's public set, with Brier-score calibration reported alongside.
- Released as v19 with weights on Hugging Face plus the training data and scripts; the older slot head performed worse.
- Frames decision models as the class accelerating since TypeSafe's Jev launch: typed options in, calibrated scores out.
Watch out
Grounded in the October 1, 2026 Strands Agents blog post captured on 2026-10-15; accuracy and latency figures are the team's own measurements, the post's later sections are truncated, and the captured text names no licence for the weights.
Questions about this entry
Short answers drawn from the linked project or page.
A small, open-source decision model from AWS's Strands Agents team: 2 billion parameters, built for fast experimentation and local development, that answers offered options and assigns scores instead of generating text.
It belongs to the same class—decision, or System One, models—and the post says the category has drawn attention since TypeSafe's Jev launch earlier in October 2026; Strands Decider is AWS's own contribution and is measured on JevBench's public set.
It takes a pretrained Qwen3.5-2B torso, removes the language-model head, and adds a pointer head of just over a million parameters that scores each option's hidden state; the torso is fine-tuned with a rank-16 LoRA adapter.
The team reports a median of around 115ms per decision on an Nvidia RTX 3090, with small tasks around 153ms on an M3 MacBook; latency grows approximately linearly with task size.
On JevBench's public set, the team reports 3rd of 33 in the 2B class for accuracy and Brier-score calibration, or 1st of 30 when just-over-2B models are excluded, with improvements across the v19 iterations.
Generate text, by design: the post says that rules it out for coding, chatbots and summarization and makes it weaker than reasoning models on complex problems, since it produces all outputs in a single parallel pass.
Two reasons given: it can be run, and even trained, on hardware you already own, keeping experimentation low-risk, and 2 billion parameters is described as something of a sweet spot for this class.
More Jev resources like this
An open-source, Jev-compatible System One model: Dex answers choice, score and yes/no questions over text in a single forward pass from an Ettin-1B encoder, and speaks the same /v1/systemone shape as Jev.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
canadaornaw/dexPython
canadaornaw/dexPythonOpen decision models from a self-improvement loop that publishes every experiment, failures included. They answer Noul, Choice and Score questions on the same POST /v1/systemone wire format as Jev, returning a calibrated probability per option in one forward pass; the 27B scores 65.64 on the Decision Index 0.3 set.
Jev Models & Reimplementations#community#decision-models#system-one
Community
Shanghua-Gao/RSI-JevHTML
Shanghua-Gao/RSI-JevHTMLOpen System One decision models (27B, 9B and 4B on Qwen chat bases, in bf16, GGUF and MLX) that answer Choice, Noul and Score questions with a probability for every option from one forward pass. The repo carries the trainer, data builders, evaluation harnesses and a record of every run.
Jev Models & Reimplementations#community#decision-models#open-weights
Community
getainode/jebadiahPython
getainode/jebadiahPythonAn open, local attempt to bring Jev-style typed decisions to images: a Qwen3-VL backbone returns numerical distributions for independent choice, score and noul questions without generating answer text.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
andrueandersoncs/visual-jevPython
andrueandersoncs/visual-jevPythonGestalt Labs' Jeff is a LoRA adapter on Qwen3-4B-Instruct-2507 that answers Jev-style choice, noul and score questions with labels and probabilities instead of prose, runs locally, and ships a train-your-own guide; on 9,730 fact-checking examples it trails Jev 1.13.0 on accuracy (0.8183 vs 0.8283).
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
A CPU port of SemIf with a web UI: JEV-CPU loads Qwen3-0.6B in float32 and reads a decision from its option logits in about a second, across eight domains from support to incident severity, with no GPU.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.