Hcompany/Holo1-3B

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:May 20, 2025License:otherArchitecture:Transformer0.1K Featherless Exclusive Cold

Hcompany/Holo1-3B is a 3 billion parameter Action Vision-Language Model (VLM) developed by HCompany, fine-tuned from Qwen2.5-VL-3B-Instruct, with a 32768 token context length. It is specifically designed for web agent systems to interact with web interfaces like a human user, excelling in UI localization and achieving state-of-the-art performance on the WebVoyager benchmark. Holo1-3B offers a strong accuracy-to-cost tradeoff for web automation tasks, making it suitable for applications requiring precise visual understanding and interaction within digital environments.

Loading preview...

Holo1-3B: Action Vision-Language Model for Web Agents

Holo1-3B, developed by HCompany, is a 3 billion parameter Action Vision-Language Model (VLM) specifically engineered for web agent systems. It enables agents to understand and interact with web interfaces in a human-like manner, acting as a policy, localizer, or validator within a broader agentic architecture. Fine-tuned from Qwen2.5-VL-3B-Instruct, Holo1-3B is optimized for both accuracy and cost-efficiency, making it a robust open-source alternative for web automation.

Key Capabilities

  • Web Interaction: Designed to interact with web interfaces, supporting tasks like planning, driving agent behavior, and precise UI interactions.
  • State-of-the-Art Performance: Achieves top performance on the WebVoyager benchmark, offering an excellent accuracy/cost tradeoff for real-world web tasks.
  • UI Localization: Excels in identifying precise coordinates on user interfaces for interaction, validated across multiple benchmarks including Screenspot, Screenspot-V2, Screenspot-Pro, GroundUI-Web, and WebClick.
  • Cost-Efficient: When integrated into the Surfer-H system, Holo1-3B-powered agents demonstrate a strong balance of accuracy and cost, outperforming larger models in cost-efficiency for web automation.

Good for

  • Developing web agents and automation systems that require human-like interaction with UIs.
  • Applications needing precise UI element localization and interaction.
  • Building cost-effective solutions for web task automation, such as data retrieval, form filling, or scheduling.
  • Research and development in vision-language models for agentic systems.

Finetunes
1 models