Strands Decider 2B
AWS Strands Labs' 2B decision model: Qwen3.5-2B with a pointer head scoring options and calibrated confidence, plus a CLI.
Overview
Strands Decider is a small, fast decision model from AWS Strands Labs for agentic workflows built with the Strands Agents SDK. It picks between options or rates on a scale, sitting between LLMs and traditional classifiers: general-purpose without training a classifier, and faster than an LLM. Per the list summary it is Qwen3.5-2B with the LM head replaced by a pointer head of about a million parameters that scores each option, plus a rank-16 LoRA, and the weights, training data and scripts are public. The launch post uses it to check a Strands agent's tool calls before they run.
Every decision carries a confidence; the README reports that on unseen short classification tasks, answers at confidence 0.9 or more are right about 95% of the time, and advises confirming or asking a person below that. A CLI (strands-decider ask) handles choice, noul and score questions from the command line, with CPU, CUDA, MPS and MLX devices, and an optional vision extra accepts images. The README includes a results table across JevBench, held-out tasks and Decision Index with memory and latency.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which team should handle this? | billing · sales · retail |
| Yes / no | Does this convey urgency? | — |
| Score | How positive is this text (0 to 1)? | — |
How it works
- pip install strands-decider
- Ask choice, noul or score questions with state via CLI or API
- Read probabilities and confidence
- Act at 0.9+ confidence; confirm or escalate below
Strengths
- About 95% accuracy at 0.9+ confidence on unseen short tasks per the README
- Small enough for CPU or Apple Silicon
- Open weights, data and scripts
Caveats
- Below 0.9 confidence answers need confirmation
- MLX extra pending release
Getting started
pip install strands-decider, then strands-decider ask StrandsAgents/strands-decider-2B-qwen3.5-v1-2610 --state "..." --choice "Question?=a,b,c".At a glance
- Type
- Model
- Status
- beta
- Language
- Python
- Runtime
- PyTorch/MLX CLI and library
- Has code
- yes
- Stars
- 612
Use cases and integrations
How curated lists describe it
AWS Strands Labs' decision model: Qwen3.5-2B with the LM head replaced by a pointer head of about a million parameters that scores each option, plus a rank-16 LoRA. Weights, training data, and scripts are public; the [launch post](https://strandsagents.com/blog/introducing-strands-decider) uses it to check a Strands agent's tool calls before they run.
AnotiaWang/awesome-decision-modelsA small, fast decision model, or system one model, for agentic workflows. Pick between options or rate on a scale faster than an LLM, with a calibrated confidence on every decision.
wh000wh000/awesome-jev-live
Frequently asked questions
How reliable is its confidence?
The README reports answers at 0.9+ confidence are right about 95% of the time on unseen short tasks.
How is it built?
Qwen3.5-2B with a small pointer head and a LoRA, per the list summary.
Does it need a key?
No; it runs locally.
Related projects
spark-s1: local Jev-inspired decision models on Qwen3.5-4B for NVIDIA DGX Spark, trained on code- and solver-labelled decision rows.
ModelresearchPythonOpen 0.8B and 2B Qwen3.5 decision models with a local systemone-compatible server, MCP tools, agent skills and a Claude Code guard.
ServicebetaPythonMangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.
ProjectproductionPython- blink★ 22
Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.
ServicebetaC - Foq★ 11
Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.
ProjectbetaPython Runs the open Laya decision model entirely in a browser tab with WebGPU and ONNX Runtime Web: no server, no API key.
ServicebetaTypeScript