Hcompany/Holo1.5-3B

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Sep 11, 2025License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

Hcompany/Holo1.5-3B is a 3 billion parameter Vision Language Model (VLM) developed by H Company, fine-tuned from Qwen/Qwen2.5-VL-3B-Instruct, specifically designed for Computer Use (CU) agents. This model excels at user interface (UI) localization and UI-based question answering across web, computer, and mobile environments, supporting high-resolution inputs up to 3840x2160 pixels. It provides a strong foundation for automating complex digital tasks by enabling AI systems to interact with real applications.

Loading preview...

Holo1.5-3B: Foundational Model for Computer Use Agents

Hcompany/Holo1.5-3B is a 3 billion parameter Vision Language Model (VLM) developed by H Company, specifically engineered to power Computer Use (CU) agents. These agents interact with real applications (web, desktop, mobile) by navigating interfaces, manipulating elements, and answering questions about content. Holo1.5-3B is fine-tuned from Qwen/Qwen2.5-VL-3B-Instruct and is part of the Holo1.5 family, which focuses on advancing automation and productivity tools.

Key Capabilities

  • UI Localization: Accurately identifies the positions of UI elements (buttons, text boxes, images) across diverse interfaces. The 3B model remains competitive with previous 7B models in this area, establishing a strong balance between size and accuracy.
  • UI-based Question Answering (QA): Comprehends UI structure and functionality to answer questions, enabling effective interaction with applications. It shows significant improvements over its base model, Qwen2.5-VL-3B, in various QA benchmarks.
  • High-Resolution Processing: Natively supports high-resolution inputs up to 3840 × 2160 pixels, crucial for interpreting complex screens.
  • Robust Training: Utilizes a multi-stage training pipeline with proprietary data, large-scale synthetic data, and human-annotated samples, including supervised fine-tuning and online reinforcement learning (GRPO).

Good For

  • Developers building Computer Use (CU) agents that need to interact with graphical user interfaces.
  • Applications requiring precise UI element localization for automation tasks.
  • Systems needing to understand and reason about screen content through question answering.
  • Projects where a smaller, efficient VLM is preferred without significant compromise on UI understanding and localization performance.

Finetunes
1 models