Jev directory
CommunityJev Models & Reimplementations
Strands-Decider 2B Qwen3.5 v1 2610
The first v1 checkpoint of the Strands decider line from the Strands Agents team: a LoRA adapter and readout head on Qwen3.5-2B-Base that answers typed decisions with calibrated confidence, reporting 0.7792 JevBench public accuracy (180/231).
- Reading time
- 2 min
- Published by
- Community
- Author
- Strands Agents
- Last verified
- Oct 2026
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- LoRA adapter plus a small readout head over Qwen/Qwen3.5-2B-Base, answering noul, choice and score questions with calibrated confidence.
- Model-index: JevBench public at 4096-token serving scores 0.7792 (180/231); held-out short tasks 0.653 (n=6000).
- v1 trained on a 246,678-row mix, about twice v21's, adding code and rule-application rows and dropping 20,000 rows with restricted commercial terms.
- Distills output distributions of google/gemma-4-31B-it where the teacher agrees with the gold label, anchored by a KL term (weight 0.3) to hobson-v19.
- This checkpoint averages three training runs (seeds 0, 1, 2); Apache-2.0; the same files also sit at amazon/strands-decider-2B-qwen3.5-v1-2610.
Watch out
Card reviewed 2026-10-17; the card text in this digest is truncated after the v1 changes, and every accuracy figure is a publisher model-index entry rather than an independent run. The improvements are documented against hobson-v21 on the card, so comparisons across the family are self-reported.
Questions about this entry
Short answers drawn from the linked project or page.
The first v1 checkpoint in the Strands decider line: a small, fast decision model for agentic AI. It picks between sets of options and rates things on a scale, and every answer carries a calibrated confidence.
The card names model routing, tool selection, argument checking, triage, guardrails, evaluations and the rote decisions of a hybrid agent that leaves the hard ones to an LLM.
A balanced 246,678-row mix about twice v21's, with new code and rule-application rows; a larger teacher (gemma-4-31B-it outputs where it agrees with gold); a KL anchor to strands-decider-2B-hobson-v19 at weight 0.3; and this checkpoint averages three seeds.
It answers typed questions (noul, choice, score) over a state with calibrated confidence instead of generating text, which is the same shape as Jev's System One contract, though the card does not present it as a drop-in Jev API.
JevBench public 0.7792 (180/231) at 4096-token serving, held-out short tasks 0.653 (n=6000), boardgame 0.81, contractnli 0.878 and Musique around 0.88. All are model-index entries from the publisher, not third-party runs.
A LoRA adapter and small readout head on Qwen/Qwen3.5-2B-Base, released under Apache-2.0. The card notes the same files are also published at amazon/strands-decider-2B-qwen3.5-v1-2610.
The E2B and E4B checkpoints use google/gemma-4-E2B-it and E4B-it bases; this one uses Qwen3.5-2B-Base. The card reports JevBench 0.7792 here, below the E4B's 0.8095 and equal to the E2B's 0.7792.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
A pointer-network decision model from Keyush Nisar that reads one agent state document plus typed questions and returns calibrated tool-routing probabilities in a single forward pass, reporting 89.00% zero-shot tool routing.
Jev Models & Reimplementations#community#open-models#local
Community
A Strands decider v1 checkpoint: a LoRA adapter and readout head on google/gemma-4-E4B-it that answers typed decisions with calibrated confidence; its model-index reports 0.8095 JevBench public accuracy (187/231) and mixed code-bug evaluation scores.
Jev Models & Reimplementations#community#open-models#local
Community
A Strands decider v1 checkpoint: a LoRA adapter and readout head on google/gemma-4-E2B-it that answers typed decisions with calibrated confidence; its model-index reports 0.7792 JevBench public accuracy (180/231) and code-bug evals from 0.281 to 0.978.
Jev Models & Reimplementations#community#open-models#local
Community
A 2B decision model from Strands Agents for agentic workflows: a LoRA adapter on Qwen/Qwen3.5-2B-Base with a readout head that answers noul, choice and score questions with calibrated confidence, reporting 0.7619 on JevBench at 4096 context.
Jev Models & Reimplementations#community#open-models#local
Community
Open Spark Jev's Qwen3.5-4B decision model on Ollama: one forward pass returns a calibrated probability per option for allow, ask, or deny tool gating, with 4b-v6 and 4b-v8 tags carrying fitted temperatures of 1.481 and 2.5 under Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune for coding decisions: roughly 9k rows from SWE-Gym teach keep/drop file relevance, priority scoring and injection detection, reporting 0.833 accuracy on 1,150 held-out rows at 126 ms p50.
Jev Models & Reimplementations#community#open-models#local
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.