apus-ailab/APUS-OpenJev-v1-9B

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 22, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

APUS-OpenJev-v1-9B is a 9 billion parameter Qwen3.5-based decision model developed by APUS AI-LAB, optimized for browser action selection, workflow routing, and natural-language principle judgments. It features a 32768 token context length and is designed to score dynamic candidates, returning their distribution for application-specific structured workflow outputs. The model achieves 85.00% on the Frozen80 development panel, covering tasks like Browser, HelpSteer3, BoolQ, MNLI, and attribute decisions.

Loading preview...

APUS-OpenJev-v1-9B: A Qwen3.5-based Decision Model

APUS-OpenJev-v1-9B is a 9 billion parameter model built upon the Qwen3.5 architecture, specifically designed for advanced decision-making tasks. Developed by APUS AI-LAB, this model excels in areas such as browser action selection, complex workflow routing, and making judgments based on natural language principles. It leverages Qwen's robust language representations and vocabulary projection, allowing application code to construct structured workflow outputs from its decisions.

Key Capabilities

  • Dynamic Candidate Scoring: The model can score dynamic candidates provided with each request and output their probability distribution, indicating relative preference.
  • Flexible Runtime: It includes a native runtime supporting both effort=low (16 layers) and effort=high (32 layers), with the high-effort mode recommended for text generation tasks.
  • Decision-Making Performance: On the Frozen80 development panel, which includes tasks like Browser, HelpSteer3, BoolQ, MNLI, and attribute decisions, the model achieves a score of 68/80 (85.00%) at full depth.
  • Quantized Versions Available: Optimized GGUF and MLX versions are provided for various platforms, including Ollama, llama.cpp, LM Studio, and Apple Silicon Macs, maintaining strong performance.

Good For

  • Automated Browser Interactions: Selecting optimal actions within a browser environment.
  • Complex Workflow Automation: Routing and managing intricate workflows based on contextual information.
  • Natural Language Principle Judgments: Making decisions or assessments based on natural language rules or guidelines.
  • Applications Requiring Candidate Ranking: When a system needs to evaluate and rank multiple potential options based on input.

Finetunes
4 models