CommunityJev Models & Reimplementations
Gero-4B
A 4B scorer rebuilt from Qwen3-4B into a branching cross-encoder: one branch per option, no position bias, up to 256 options, and calibrated probabilities with no text to parse.
- Published by
- Community
- Author
- vixhal-baraiya
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Rebuilt without a language-model head: a single learned linear scorer reads the final token of each option branch.
- Because options are read in parallel and never see each other, reordering them changes no probability, and 2 and 256 options work the same way.
- Training runs in three stages — readout, instruction structure with shortcut auditing, then RL against a proper scoring rule.
- The name comes from Gerolamo Cardano's Liber de Ludo Aleae, the 1564 first systematic study of probability.
Watch out
The card publishes no benchmarks, accuracies, or comparisons — only mechanics, so there is no evidence to weigh. English is the training language, the top of the confidence range runs slightly optimistic, and the card says to set thresholds on your own data.
Questions about this entry
Short answers drawn from the linked project or page.
No. It returns a calibrated probability per answer with nothing to parse. Options can be a plain list or name-to-description mappings, and each option is judged against the state alone; a result near 0.5 means yes and no are about equally likely, not half true.
It is a branching cross-encoder: the state and question share a prefix, each option gets its own branch, and attention runs only one way. That removes position bias by construction and lets the option count vary up to 256 for Choice and 2–10 levels for Score.
Apache 2.0, the same as its base model. It starts from Qwen/Qwen3-4B and is rebuilt into a scorer, and the licence carries over to the rebuilt weights.
Ask each question one narrow thing a knowledgeable person could decide in seconds. When a decision depends on several factors, ask about each separately and combine the answers in code; arithmetic, lookups, and fixed business rules belong in code too.
Choice accepts 2 to 256 options, Score takes 2 to 10 ordered levels, and yes/no returns the probability of yes. Each option is judged against the state on its own and options never see one another.
No. The card publishes only mechanics - architecture, question types, confidence, and training - with no accuracies, benchmarks, or comparisons, so there is no evidence to weigh; it says to set thresholds on your own data.
Its probabilities are trained to be taken at face value, but the very top of the range runs slightly optimistic, so leave margin on thresholds close to 1. English is the training language, so test other languages on your own content.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Show HN: Ad Spend Guard gates budget before the auction, not the creative after.
Show HN: Gating ad budget with a decision modelRead the thread on Hacker NewsX: Field Advisory sends marginal readings to an agronomist instead of a spray.
This post couldn't be loaded here — read it on X.https://x.com/agriwisehq/status/2781000000000000003LinkedIn: 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 NewsMore Jev resources like this
A 0.4B Jev-compatible decision model with measured order invariance: 960 random option shuffles moved no probability by more than 1e-5, and bev-decider serve exposes /v1/systemone.
Jev Models & Reimplementations#community#open-models#local
Community
Supersonic Labs' 144.3M-parameter decision model on mmBERT-small: choice, score, and noul answers from supplied options in one pass, 73.15% on 2,000 typed decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Liquid AI's 3.12B decision model built on LFM2.5-VL: state plus typed questions in one forward pass with zero output tokens, 48.57 on Decision Index 0.2.1 and 8 ms per decision on an RTX 4090.
Jev Models & Reimplementations#community#open-models#local
Community
An untrained readout: stock Gemma-4-31B weights plus a decider config that reads option letters at the answer slot, scored #2 of 70 on Decision Index v0.2.1 at 57.33 against Jev 1.13.0's 57.91.
Jev Models & Reimplementations#community#open-models#local
Community
Nace.AI's 8B decision model on NVIDIA's Efficient-DLM diffusion backbone: a 256-dim pointer head scores every option in one pass, behind the same POST /v1/systemone API as its hosted service.
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
From the Jev guides
Original write-ups on what this entry is for.
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Hundreds of repositories carry the system-one tag, but the ecosystem is a dozen ideas in different sizes. A tour of the families, the licenses, and what a download does not include.
Jev Models & Reimplementations#open-models#hugging-face#local
Read article
From the Jev community
Posts from builders shipping with Jev right now.