Hcompany/Holo4-27B

Hugging Face
VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 24, 2026License:cc-by-nc-4.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Warm

Hcompany/Holo4-27B is a 27 billion parameter vision-language model (VLM) developed by H Company, built upon the Qwen3.8 dense architecture. Designed for Computer Use, it excels at interacting with graphical interfaces, executing code, and making tool calls. This model is specifically optimized for agentic workflows across web, desktop, and mobile environments, enabling it to interpret screenshots and perform actions like clicks and typing.

Loading preview...

Holo4-27B: A Vision-Language Model for Computer Use

Holo4-27B, developed by H Company, is a 27 billion parameter vision-language model (VLM) built on the Qwen3.8 dense architecture. It is specifically designed for interacting with computer interfaces, enabling agents to perceive screenshots and execute actions such as clicks, typing, code, and tool calls. This model is part of the Holo4 family, which also includes the MoE architecture Holo4-35B-A3B.

Key Capabilities

  • Vision-Language Understanding: Processes visual input (screenshots) to understand the state of a graphical interface.
  • Agentic Control: Integrates with the hai-agents harness to send screenshots and tool results, then executes requested actions.
  • Multi-Environment Interaction: Capable of operating across web, desktop, and mobile target environments.
  • Tool and Code Execution: Supports function calling, element localization, and document OCR, allowing for complex automated workflows.

Performance Highlights

Holo4-27B demonstrates significant improvements over its Qwen base models. On the OSWorld benchmark, it scores 85.2% at $0.08 per task. It also performs well on Agentic Task Factory, a set of business workflows involving web, desktop, and MCP tools. For transparency, H Company shares all agent trajectories in the open-source Hcompany/trajectories dataset.

When to Use Holo4-27B

This model is ideal for developers building intelligent agents that need to automate tasks requiring visual perception and interaction with various software applications. Its ability to interpret graphical interfaces and execute precise actions makes it suitable for complex automation, testing, and assistive technologies.

Finetunes
5 models