OmniJev-OneJev1.5-27B-MLX

OneJev1.5-27B is the 27B model of OmniJev's OneJev 1.5, a multimodal "System One" decision model built as a full fine-tune of Qwen3.8-27B on about 320,000 questions drawn from real agent runs, videos, images and the training splits of public decision datasets. It takes a screenshot, photo, video or plain text along with a few typed questions, such as Noul yes/no or Choice over labeled options, and returns a calibrated probability for every option in a single forward pass. On the public part of Decision Index 0.3 (37 benchmarks, 140,178 requests, run and scored with the board's own kit) it scores 63.64, fifth of the 148 models on the board. Its Arts & Human Taste score is the highest on the board. In the card's five-model class comparison it trails RSI-Jev v6.1 VL (65.6), vLLM-SR d3 (65.5), Torchcast Decision (65.1) and Decider 31B (64.3), the four highest-scoring models of 25B to 36B parameters. Its Language Understanding score is second among those five. Those comparison figures are board public-part scores as of 12 October 2026. It also reports accuracy against Jev 1.13, Jev-Omni 12B and Qwen3.8-27B in thinking mode on a held-out OneJev test set, but only as a graphic, so those numbers are not readable from the text. The card notes that Jev 1.13's figures are its published ones and that it reads text only. It is served with the qev package, whose server speaks TypeSafe's System One API and adds a media field for images and video. It is released under Apache 2.0.

Directory Structure

prithivMLmods/OmniJev-OneJev1.5-27B-MLX/
├── 4bit/
├── 8bit/
└── (root: bf16 base model mlx)

Use with mlx

Install the required library:

pip install -U mlx-vlm

Task Note: OmniJev-OneJev1.5-27B operates as a multimodal fast decision-engine for agent actions, UI decisions, and visual multiple-choice verification.

BF16 Variant (Base Weights)

The unquantized BF16 mlx weights are located directly in the root of the repository:

CLI (Terminal)

python -m mlx_vlm generate \
  --model prithivMLmods/OmniJev-OneJev1.5-27B-MLX \
  --max-tokens 256 \
  --temperature 0.0 \
  --prompt "Analyze this screenshot and determine the optimal next UI action: Choice [A: Click Submit, B: Refresh Page, C: Scroll Down]." \
  --image <path_to_image>

Python API

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

model_path = "prithivMLmods/OmniJev-OneJev1.5-27B-MLX"
model, processor = load(model_path)
config = load_config(model_path)

image = ["<path_to_image>"]
prompt = "Analyze this screenshot and determine the optimal next UI action: Choice [A: Click Submit, B: Refresh Page, C: Scroll Down]."
formatted_prompt = apply_chat_template(processor, config, prompt, num_images=len(image))

output = generate(
    model, 
    processor, 
    formatted_prompt, 
    image=image, 
    max_tokens=256, 
    temperature=0.0
)
print(output.text)

8-bit Variant

Access the 8-bit quantized files using --subfolder 8bit:

CLI (Terminal)

python -m mlx_vlm generate \
  --model prithivMLmods/OmniJev-OneJev1.5-27B-MLX \
  --subfolder 8bit \
  --max-tokens 256 \
  --temperature 0.0 \
  --prompt "Analyze this screenshot and determine the optimal next UI action: Choice [A: Click Submit, B: Refresh Page, C: Scroll Down]." \
  --image <path_to_image>

Python API

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

model_path = "prithivMLmods/OmniJev-OneJev1.5-27B-MLX"
model, processor = load(model_path, subfolder="8bit")
config = load_config(model_path, subfolder="8bit")

image = ["<path_to_image>"]
prompt = "Analyze this screenshot and determine the optimal next UI action: Choice [A: Click Submit, B: Refresh Page, C: Scroll Down]."
formatted_prompt = apply_chat_template(processor, config, prompt, num_images=len(image))

output = generate(
    model, 
    processor, 
    formatted_prompt, 
    image=image, 
    max_tokens=256, 
    temperature=0.0
)
print(output.text)

4-bit Variant

Access the 4-bit quantized files using --subfolder 4bit:

CLI (Terminal)

python -m mlx_vlm generate \
  --model prithivMLmods/OmniJev-OneJev1.5-27B-MLX \
  --subfolder 4bit \
  --max-tokens 256 \
  --temperature 0.0 \
  --prompt "Analyze this screenshot and determine the optimal next UI action: Choice [A: Click Submit, B: Refresh Page, C: Scroll Down]." \
  --image <path_to_image>

Python API

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

model_path = "prithivMLmods/OmniJev-OneJev1.5-27B-MLX"
model, processor = load(model_path, subfolder="4bit")
config = load_config(model_path, subfolder="4bit")

image = ["<path_to_image>"]
prompt = "Analyze this screenshot and determine the optimal next UI action: Choice [A: Click Submit, B: Refresh Page, C: Scroll Down]."
formatted_prompt = apply_chat_template(processor, config, prompt, num_images=len(image))

output = generate(
    model, 
    processor, 
    formatted_prompt, 
    image=image, 
    max_tokens=256, 
    temperature=0.0
)
print(output.text)

License and Attribution

This model is based on and/or incorporates the following open-source projects and models:

• OneJev1.5-27B: https://huggingface.co/OmniJev/OneJev1.5-27B
• mlx-vlm: https://github.com/Blaizzy/mlx-vlm
• MLX: https://github.com/ml-explore/mlx

This model is released under the Apache License 2.0.