Strands Decider 2B vs Jev

Strands Decider 2B is AWS's open-source decision model, released by Strands Labs on 1 October 2026. At ~1.9B parameters it's the small, local end of the category Jev defined: it reads text plus typed questions (yes/no, pick one, or score) and returns each answer with a confidence score in one forward pass — running on your own laptop in about 115ms. Here's the honest comparison with hosted Jev, and when a tiny local model beats a managed API.

Strands Decider 2B and Jev answer the same kind of question — a bounded decision with a probability attached — but from opposite corners. Strands Decider is a tiny, fully open model you run locally for near-zero marginal cost; Jev is a hosted, calibrated model you call. If you were weighing them, the axis is local-and-open vs hosted-and-calibrated, plus how much accuracy a 1.9B model can give up.

What Strands Decider 2B is

Strands Labs is the experimental arm of AWS's open-source Strands Agents project. Strands Decider 2B is a 1.9B-parameter decision model based on Qwen3.5-2B-Base with a LoRA (rank 16) adapter and a pointer head. It reads a piece of text plus typed questions — yes/no, pick-one, or a score on a scale — and returns each answer with a confidence score in a single forward pass, rather than generating text. AWS positions it for ticket triage, routing and guardrails, optimised for fast experimentation and local development. Everything ships under Apache-2.0, including the full training data and scripts, so you can reproduce or retrain it end to end.

Specs, side by side

The Strands Decider column uses AWS's published figures; the Jev column uses TypeSafe's. Every number here is vendor-reported.

The real trade-off: 1.9B local vs hosted and calibrated

A 1.9B model that runs on your laptop in ~115ms is a genuinely different proposition from a hosted API: no per-call price, no network hop, nothing leaves the machine. The cost is accuracy headroom and the ops of running it. On the public JevBench set AWS reports its reference checkpoint at 0.762 — strong for the size, but JevBench uses Jev as its reference answers, so read that as 'close to Jev on these tasks', not a clean ranking. If the decision is load-bearing, a bigger model's calibration may earn its keep; if you're prototyping, triaging at volume, or need everything on-device, a tiny open model is hard to beat on cost.

When to pick which

• Reach for Strands Decider 2B when you want a small, fully open model to run locally — offline or on-device, zero marginal cost per decision, full training data to retrain, and a natural fit if you're already building on Strands Agents.
• Reach for Jev when you want a hosted, RLCD-calibrated model with pinned versions and honest confidence, no inference to run, headroom on harder decisions, and a free browser playground to try one call before you integrate.
• They compose: prototype or triage locally on Strands Decider and escalate the uncertain tail to hosted Jev — both speak the same state-plus-typed-questions shape.

Make one real decision on Jev

The fastest way to judge any decision model is to run your own decision through it. Jev's playground does it in the browser with no signup, and a jv_live_ key points your code at the hosted endpoint — the same shape a local model like Strands Decider expects:

// Hosted, calibrated decision — the same text + typed-questions shape
// a local decision model like Strands Decider 2B expects.
const res = await fetch("https://jevtypesafeai.com/api/v1/decide", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    Authorization: `Bearer ${process.env.JEV_KEY}`, // jv_live_...
  },
  body: JSON.stringify({
    state: { ticket: ticket.body },
    questions: {
      triage: { type: "choice", instructions: "First-pass triage.",
                criteria: { resolve: "answerable now", escalate: "needs a human", spam: "not a real request" } },
      urgent: { type: "noul", instructions: "Is it urgent?" },
    },
  }),
});
const { answers } = await res.json();
if (answers.triage.choice === "escalate") handoff(ticket, answers.urgent.noul);

• MarkTechPost — AWS Strands Labs releases Strands Decider 2B — 1.9B Qwen3.5-2B + LoRA + pointer head, ~115ms, Apache-2.0 with full training materials (vendor-reported)
• The New Stack — AWS launches a local answer to TypeSafe's Jev — Positioning vs Jev, local-development focus, JevBench scores

FAQ

See also: Jev alternatives · JevBench · Run Jev-like models locally · What is a decision model · Playground