What hardware does OmniJev need?

Requirements list Python 3.9+ for the SDK and HTTP service, Python 3.12+ for the workbench, and a validated NVIDIA A800 80GB with about 25 GB of GPU memory for the model service.

What does the model actually return?

One scored choice among the options supplied with the request, plus the option probabilities. One forward pass scores candidates and no answer token is generated.

Does OmniJev generate a text answer?

No. One forward pass scores the options supplied with the request and returns their probabilities; the technical report says no answer token is generated. The probability bars are softmax over option-letter logprobs, not calibrated confidence.

What happens when the model cannot see what it needs?

It abstains rather than guessing. In the twelve probe families, the four cases with no image, an unobservable property, a blank frame or a color outside the candidate set all return insufficient evidence, which the finite-choice interface is built to make explicit.

How does OmniJev compare with the two 4B baselines?

On 923 held-out image questions across three official interfaces it scores 76.5% total against 72.3% for NeoHorse-Jev-4B and 68.7% for tinnel OmniJev-4B. The report notes this is a 9B model against two 4B models on one fixed suite.

Which model files do I need to download and place?

Place Qwen/Qwen3.5-9B at models/Qwen3.5-9B and the v4 lora.pt, head.pt and config.json at models/runs/qwen35-rlcd-v4. The 9B backbone is not redistributed in the repository, and the model files total about 26.1GB.

Are the robot trajectory replays end-to-end visual planning?

No. The released replays use skill mode with privileged simulator state, where the model chooses among preset skills and a program executes the IK trajectory. Dual-camera vision input is a separate mode with separately reported results.

CommunityJev Models & Reimplementations

OmniJev

Multimodal finite-choice decision interface and MuJoCo robotic-arm workbench: dual-camera images and state go to a self-hosted Qwen3.5-9B with a corrective head, which returns one typed next-skill choice without generating text.

See highlights

Open github.com

Stars: 21
Language: Python
License: MIT
Published by: Community
Last verified: Oct 2026

shapsider/OmniJev

User repository on GitHub

View on GitHub

OmniJev — multimodal finite-choice decision interface and MuJoCo embodied workbench: trajectory replays, decision probes, benchmark panels, 60s walkthrough.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

OmniJev is a multimodal research project for robot decision-making, providing a finite-choice decision SDK, an HTTP service, and a MuJoCo robotic arm workbench.

Quoted from the project's README. Read the full README.

Tagscommunity robotics multimodal mujoco vision

Works forJev for Realtime control Jev for Decision gating

Used inJev for Robotics & Hardware

Languages: Python
Capabilities: Open sourceSelf-hostedRuns locally

Highlights

Current model is Qwen3.5-9B plus a v4 corrective head; weights are the LoRA and residual head at tzcfly/OmniJev-Qwen3.5-9B-v4, around 26.1 GB
On 923 held-out image questions across three official interfaces it scores 76.5% total versus 72.3% and 68.7% for two 4B baselines
Released episodes finish transfer, stack and barrier tasks in 13 decisions and 39 output tokens each
208 of 208 non-audio probe requests were valid; unobservable inputs return insufficient evidence instead of a guess
Four public benchmark pilots report paired accuracy with direct short answer while cutting decision latency 6.8-13.6x and total tokens 51-86%

Watch out

MIT license. Preview quality: skill-mode results use privileged simulator state and are not end-to-end visual planning; audio is wired in the client but rejected by the current backend.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Hacker News: Pick Verify: a single decision at the warehouse pick station catches confusable picks.

Show HN: Verifying robot picks with a decision modelRead the thread on Hacker News

LinkedIn: What multimodal decision models actually change, and what they do not.

LinkedInJevDirectory.org

Hacker News: A one-token logprob wrapper that also classifies webcam frames.

allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker News

X: A dual-arm robot with Jev in the middle layer

More Jev resources like this

AutoTrust JEV-27B-VL

The vision sibling: the same text-trained System 1 head on Qwen3.8-27B's image encoder, taking images zero-shot, ranked first of 20 vision decision models on the community Jev Decision Index 0.3.

Jev Models & Reimplementations#community#open-models#local

Community

intern-decision-mlx

Serves Shanghai AI Lab's Intern-Decision-0.8B vision decision model on Apple Silicon behind a /v1/systemone-shaped endpoint: screenshots plus typed questions in, calibrated Choice, Score and Noul probabilities out, for thresholding and escalation.

Jev Models & Reimplementations#community#apple-silicon#local

Community

dex0shubham/intern-decision-mlxPython

Jev Robot

MuJoCo simulation where a local decider-2b model picks the next skill for an AgileX PiPER arm through a Jev-style typed-choice interface while the red cube target is pushed to new positions mid-run. Deterministic checks may reject a chosen skill but never substitute another.

Jev Demos & Experiments#community#robotics#simulation

Community

Hu-xiao-max/jev_robotPython

jevos

1.3k

A local yes/no decision model that speaks Jev's wire format on a laptop CPU, refusing Choice and Score until those question types ship.

Jev Models & Reimplementations#community#python#open-models

Community

feder-cr/jevC++

Valen

738

A multimodal decision model that scores text, image, and video candidates with a shared head on Qwen3.5, plus SFT and experimental RLCD training code.

Jev Models & Reimplementations#community#python#open-models

Community

Liuziyu77/ValenPython

OpenDecision

An Apache-2.0 decision engine with a TypeSafe-compatible endpoint, document answers with source passages, and a ViZDoom demo that picks actions from structured game state.

Jev Models & Reimplementations#community#python#local

Community

deepanwadhwa/OpenDecisionPython

From the Jev guides

Original write-ups on what this entry is for.

Multimodal Decision Models: What Clef and d1 Change

Oct 12, 2026

Jev takes text only; Clef-omni adds image, audio, and video input and Liquid d1 adds images, each keeping the same typed-answer shape. Where that matters, where it does not, and the costs the vendors underplay.

Jev Models & Reimplementations#multimodal#open-models#decision-model

Read article

The Decision-Model Price War

Oct 11, 2026

One model in September, a dozen by October. Clef, Liquid d1, Perplexity, and OpenAI all ship the same typed-answer shape now. How to read the rival benchmarks and pick on your own task.

Jev Models & Reimplementations#decision-model#benchmarks#pricing

Read article

What Is Actually on Hugging Face

Oct 8, 2026

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

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai