Jev directory
CommunityJev Models & Reimplementations
Decision-2.0-Nox-4B
The 4B model in vLLM Semantic Router's Decision 2.0 family: a 4.21B fine-tune of Qwen3.5-4B-Base that answers choice, yes/no and score questions in one forward pass, with a self-reported JevArena 63.6 and a 12.9 ms median per request.
- Reading time
- 2 min
- Published by
- Community
- Author
- vLLM Semantic Router
- Last verified
- Oct 2026
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- 4.21B parameters with a 16,384-token context, Apache-2.0, fine-tuned from Qwen/Qwen3.5-4B-Base.
- Card reports JevArena 63.6, statistically level with Decider 4B's 61.9 and 7.1 above Decision 1.0 Nox.
- Jev Decision Index 43.8, up 9.4 from Decision 1.0 Nox; human-labelled transfer 52.3.
- Median 12.9 ms per single-question request on a single GPU, with choice, yes/no and score in one forward pass.
- Quickstart uses transformers 5.17 or newer with trust_remote_code=True and model.system_one(state=..., questions=...).
Quickstart
python
import json
from transformers import AutoModel
model = AutoModel.from_pretrained("vllm-sr/Decision-2.0-Nox-4B", trust_remote_code=True)
result = model.system_one(
state="The order arrived damaged yesterday. The customer has a receipt and asks for a replacement today.",
questions={
"route": {"type": "choice", "instructions": "Which team should handle this request?",
"criteria": {"returns": "Refunds, replacements and damaged deliveries", "billing": "Payments, invoices and charges"}},
"receipt": {"type": "noul", "instructions": "Does the customer have a receipt?"}},
)
print(json.dumps(result["answers"], indent=2))Watch out
Card reviewed 2026-10-17; the JevArena, Jev Decision Index, transfer and 12.9 ms figures are all self-reported by the vLLM Semantic Router team, and competitor rows come from a public board snapshot dated 2026-09-28. The card names no GPU behind the latency median.
Questions about this entry
Short answers drawn from the linked project or page.
The 4B model of Decision 2.0, the decision-model family from vLLM Semantic Router and a fine-tune of Qwen/Qwen3.5-4B-Base. Given an input plus choice, yes/no and score questions, it returns a probability for every answer without generating text.
It answers typed questions about a state in one forward pass and returns probabilities rather than generated text, which is the System One shape. The card's example asks a routing choice question, a noul receipt question and an ordinal urgency question together.
Self-reported: JevArena 63.6, human-labelled transfer 52.3 and Jev Decision Index 43.8, described as statistically level with Decider 4B's 61.9 and ahead of Decision 1.0 Nox by 7.1 and 9.4 points respectively.
A median of 12.9 ms per single-question request on a single GPU, with all of a request's questions answered in one forward pass. The card does not name the GPU or the request mix behind that median.
4.21B parameters and a 16,384-token context. It carries the zero-shot-classification pipeline tag, loads through transformers with trust_remote_code=True, and is released under Apache-2.0.
The quickstart installs transformers 5.17 or newer with torch and safetensors, loads the model with AutoModel.from_pretrained(..., trust_remote_code=True) and calls model.system_one(state=..., questions=...); a transformers pipeline('decision', ...) call is shown as an alternative.
Against Decision 1.0 Nox (56.5 JevArena, 34.4 Decision Index) it gains 7.1 and 9.4 points. Decider 4B scores 61.9 and Jet v6.2 60.4 on JevArena; competitor rows are a public board snapshot dated 2026-09-28.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: ASTGL replay chose local Laya routing over hosted Jev on one Mac Studio setup.
Jmeg8r15 pointsDoes an open-weight decision model beat a hosted one? Jev vs. LayaRead the thread on Hacker NewsReddit: A walkthrough of routing between models with Jev.
X: Standard One SH (StandardThinking)
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
More Jev resources like this
The 9B entry in vLLM Semantic Router's Decision 2.0 family: a 7.94B decision model with a 16,384-token context that answers choice, yes/no and score questions in one pass, reporting 68.1 JevArena and 18.4 ms median single-question latency.
Jev Models & Reimplementations#community#open-models#local
Community
The 0.6B model in vLLM Semantic Router's Decision 2.0 family: a 0.60B fine-tune of Qwen3-0.6B-Base with an 8,192-token context that returns probabilities for choice, yes/no and score questions from one forward pass.
Jev Models & Reimplementations#community#open-models#local
Community
The 2B model in vLLM Semantic Router's Decision 2.0 family: a 1.88B fine-tune of the earlier Decision-1.0-Sol-2B checkpoint with a 16,384-token context, reporting JevArena 52.1 and a 7.2 ms median per request.
Jev Models & Reimplementations#community#open-models#local
Community
The 0.8B model in vLLM Semantic Router's Decision 2.0 family: a 0.75B fine-tune of Decision-1.0-Eos-0.8B with a 16,384-token context, reporting JevArena 53.9 and a 6.0 ms median per single-question request.
Jev Models & Reimplementations#community#open-models#local
Community
The vLLM Semantic Router project's 29.37B decision model on Qwen3.8-27B: a probability per supplied option with zero generated tokens, 74.0 on JevArena and 56.5 on Jev Decision Index 0.2.1.
Jev Models & Reimplementations#community#open-models#local
Community
A small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
Jev Models & Reimplementations#community#open-models#multimodal
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.