OmniJev/OneJev-27B

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 27, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

OmniJev/OneJev-27B is a 27 billion parameter multimodal System One decision model developed by OmniJev. It is a full fine-tune of Qwen/Qwen3.8-27B, designed to process screenshots, photos, videos, and plain text inputs. The model returns calibrated probabilities for multiple options in a single forward pass, excelling at decision-making tasks based on diverse inputs. It demonstrates high accuracy on unseen questions and offers efficient inference speeds, making it suitable for real-time agentic applications.

Loading preview...

Overview

OmniJev/OneJev-27B is a 27 billion parameter multimodal System One decision model, developed by OmniJev. It is a full fine-tune of the Qwen3.8-27B base model, trained on 99,193 questions derived from real agent runs, videos, and images. This model is designed to take various inputs, including screenshots, photos, videos, and plain text, and provide calibrated probabilities for different options in a single forward pass.

Key Capabilities

  • Multimodal Input Processing: Handles images (screenshots, photos) and video, alongside plain text queries.
  • System One Decision Making: Provides rapid, calibrated probabilistic outputs for decision-making tasks.
  • High Accuracy: Achieves strong performance on a test set of questions not seen during training, outperforming Jev 1.13 and Qwen3.8-27B thinking on various metrics.
  • Efficient Inference: Capable of answering 1 question in 189 ms and 10 questions in 324 ms (32.4 ms per question) on an H200 GPU with a 1280x720 screenshot.
  • API Integration: Supports TypeSafe's System One API with an added media field for multimodal inputs.
  • Flexible Deployment: Can be run with PyTorch or via llama.cpp using GGUF builds, with image support in both, and video support requiring the PyTorch server.

Good For

  • Agentic Applications: Ideal for scenarios requiring quick, probabilistic decisions based on visual and textual information.
  • Automated Task Execution: Suitable for tasks like paying invoices or determining next actions in a workflow, as demonstrated in examples.
  • Real-time Decision Support: Its low latency makes it effective for applications needing rapid responses to complex, multimodal inputs.

Finetunes
3 models