apus-ailab/APUS-OpenJev-v1-4B

Hugging Face
VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 22, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

APUS-OpenJev-v1-4B by APUS AI-LAB is a 4.5 billion parameter Qwen3.5-based decision model specifically designed for browser action selection, workflow routing, and natural-language principle judgments. It excels at scoring dynamic candidates supplied with each request and returning their distribution, making it suitable for applications requiring structured decision-making from natural language inputs. The model features a 32768-token context length and includes a native runtime supporting both low and high effort modes for varied performance needs.

Loading preview...

APUS-OpenJev-v1-4B: A Qwen3.5-based Decision Model

APUS-OpenJev-v1-4B is a 4.5 billion parameter model developed by APUS AI-LAB, built upon the Qwen3.5 architecture. It is specifically engineered for advanced decision-making tasks, particularly in environments requiring the interpretation of natural language for structured outputs.

Key Capabilities

  • Dynamic Candidate Scoring: The model can score dynamic candidates provided with each request, outputting their probability distribution. This is crucial for applications like browser action selection and workflow routing.
  • Flexible Runtime: It includes a native runtime that supports two effort levels: 'low' (16 layers) for efficiency and 'high' (32 layers) for enhanced performance, especially in text generation scenarios.
  • Qwen Language Representation Reuse: Leverages Qwen's language representations and vocabulary projection, allowing application code to assemble decisions into structured workflow outputs.
  • Evaluation Performance: The merged model scores 66/80 (82.50%) on the Frozen80 development panel, which covers tasks like Browser, HelpSteer3, BoolQ, MNLI, and attribute decisions.

Good for

  • Automated Browser Interactions: Ideal for systems that need to select actions within a browser based on natural language instructions.
  • Workflow Automation: Suitable for routing and managing complex workflows by making principle judgments from natural language.
  • Structured Decision-Making: Applications requiring the model to evaluate multiple options and provide a probabilistic preference for each.
  • Integration with Qwen Ecosystem: Benefits from reusing Qwen's linguistic foundation, simplifying integration for existing Qwen-based projects.

Finetunes
4 models