Surfer H

What is Surfer H?

Surfer H is a cost-efficient web agent designed to automate web-based tasks. It leverages open weights and is powered by Holo1, a family of Visual Language Models (VLMs), enabling it to interact with web User Interfaces (UIs) much like a human user would. This allows Surfer H to see what's on the screen, decide what actions to take, interact with UIs, and determine when a task is complete.

How does Surfer H work?

Surfer H is built with a modular design consisting of three primary components:

• Policy Model: This component plans, decides, and drives the agent's behavior, determining the steps necessary to achieve the desired outcome.
• Localizer Model: This model interprets visual UIs, allowing the agent to precisely interact with web elements.
• Validator Model: This component validates whether the answer is correct and complete, ensuring the agent provides accurate results.

Surfer H operates by thinking before acting, taking notes, and retrying if its initial attempt is unsuccessful. The agent's modular architecture also allows for the use of different models for each component, providing flexibility in balancing accuracy, speed, and cost.

Key Features and Benefits

• Cost-Efficiency: Powered by Holo1, Surfer H offers a strong balance between accuracy and cost, delivering high performance at a fraction of the cost of other agents.
• Flexibility and Modularity: Its modular design allows for the use of different models for each component, enabling customization based on specific task requirements.
• Browser-Based Operation: Surfer H operates directly through the browser, eliminating the need for custom APIs or wrappers.
• State-of-the-Art UI Localization: Holo1's advanced UI localization capabilities enable Surfer H to accurately identify and interact with web elements.
• WebVoyager Benchmark Performance: Surfer H demonstrates exceptional performance on the WebVoyager benchmark, completing a wide range of real-world web tasks with high accuracy.

How to use Surfer H?

While specific usage instructions aren't detailed in the provided content, Surfer H is designed to be a general-purpose web automation system. Example use cases include:

• Job board monitoring for developer roles
• Gear comparison for fitness products
• Competitor pricing research
• Competitive landing page analysis
• Trend scouting for newsletters
• Last-minute hotel searches
• Collector search automation (e.g., tracking Pokémon card listings)
• Web search for financial reports

Who is Surfer H for?

Surfer H is ideal for businesses and individuals looking to automate web-based tasks, reduce costs, and improve efficiency. It's especially useful for:

• Businesses: Automating competitive research, data collection, and other repetitive tasks.
• Researchers: Gathering data from the web for analysis and insights.
• Developers: Building web applications and automating testing.
• Anyone: Who wants to save time and effort by automating web-based tasks.

Why choose Surfer H?

Surfer H stands out due to its combination of cost-efficiency, flexibility, and accuracy. Powered by Holo1, it offers a powerful and versatile solution for web automation, making it an excellent choice for anyone looking to streamline their web-based workflows.

Holo1: State-of-the-Art UI Localization

A key skill for the real-world utility of our VLMs within agents is localization: the ability to identify precise coordinates on a user interface (UI) to interact with, to complete a task, or follow an instruction. To assess this capability, we evaluated our Holo1 models on several established localization benchmarks, including Screenspot, Screenspot-V2, Screenspot-Pro, GroundUI-Web.

Holo1 significantly outperforms prior models like Qwen2.5-VL, UI-TARS, and UGround across these benchmarks:

-Holo1-3B: 73.6% average localization accuracy, beating other 3B and even some 7B models

-Holo1-7B: 76.2%, the highest small-size model overall

To support the community, we're also releasing Web Click, a new benchmark for UI Grounding that better reflects how humans really use the web. It includes 1,639 screenshots and instruction-label pairs from over 100 websites, designed to challenge existing VLMs.

Open Weights for Transparency and Collective Progress

H Company believes that open weights are more than just a philosophy, they are a practical tool to accelerate experimentation, transparency, and collective progress. By providing open access to the weights of Holo1, they empower the community to build upon their work and create even better agents.