Skip to main content
We’re constantly improving our foundational models and pushing the boundaries of what they’re capable of accomplishing when powering our agents. This page provides updates and announcements on our Holo models.
​
11/13/2025

​
Discover Holo2

Cost-Efficient Models for Cross-Platform Computer-Use AgentsHolo2 is our most powerful and lightweight model series for localization and navigation to date. Holo2 reaches state-of-the-art performance for UI localization, surpassing our previous Holo1.5 models of the same size and outperforming all models on the competitive ScreenSpot-Pro benchmark. When powered by Holo2, our Surfer 2 agent achieves competitive performance on major web, mobile, and computer benchmarks, at a fraction of the cost of our previous version.All our models are open weights and accessible on Hugging Face.For more information, check out our blog post or view our guides.
​
09/15/2025

​
Introducing Holo1.5

Open Foundation Models for Computer Use AgentsSince the release of Holo1, we’ve been hard at work improving our models on foundational capabilities for Computer Use agents. As of September 15 2025, we are open-sourcing the Holo1.5 series of models in 3B, 7B, and new 72B sizes. Holo1.5 adds a massive 10%+ accuracy boost compared to Holo1 for all sizes and sets a new state-of-the-art for Computer Use localization models. It also delivers strong performance on user interface (UI) understanding and question answering. All our models are open weights and accessible on Hugging Face.For more information, check out our blog post or view our guides.
​
06/03/2025

​
Holo1: New family of GUI automation VLMs powering GUI agent Surfer-H

Holo1 is the first family of open-source Action VLMs designed specifically for deep web UI understanding and precise localization. The family includes Holo1-3B and Holo1-7B models, with the latter achieving 76.2% average accuracy on common UI localization benchmarks—the highest among small-size models. H Company has released these models with open-source on Hugging Face, along with the WebClick benchmark containing 1,639 human-like UI tasks.For more information, check out our blog post or view our guides.