EN
Bite Dance has recently released Utars 1.5, a significant upgrade to its vision language agent. This innovative model treats your screen as a single image, allowing it to read, reason, and manipulate content directly. Unlike traditional methods that rely on DOM trees or external tools, Utars 1.5 can interpret a screenshot, understand the layout, and execute tasks using plain language, acting as if a real user is in control.
The latest version of Utars builds on its predecessor by integrating a more robust architecture. It features three model sizes: a lightweight 2 billion parameter model, a mid-range 7 billion model, and a powerful 72 billion variant. This upgrade includes direct preference optimization across extensive training data, enabling the model to see, reason, and act in a single pass, enhancing its performance in GUI automation and various workflows.
Utars 1.5 introduces a sophisticated perception system that analyzes various graphical user interfaces, including websites, Windows applications, and mobile UIs. The model synthesizes multiple types of perception data, allowing it to recognize elements and their functions accurately. Additionally, it features a unified action space that includes common commands like click, drag, and scroll, as well as desktop and mobile-specific actions, enabling it to perform tasks seamlessly.
One of the standout features of Utars 1.5 is its reasoning capability. The model distinguishes between intuitive and deliberate thinking processes, allowing it to break down tasks and learn from errors. By analyzing millions of GUI tutorials and action traces, it has developed a method for reasoning that includes an inner monologue before executing actions, enhancing its decision-making process.
In performance benchmarks, Utars 1.5 has demonstrated impressive results, achieving a 42.5% success rate in complex tasks, outperforming previous models like OpenAI's operator. Its ability to handle multi-step tasks and adapt to different environments, such as desktop and mobile UIs, showcases its versatility and effectiveness in real-world applications.
Bite Dance has made Utars 1.5 accessible to the broader community by releasing its model weights and training scripts under an open-source license. This allows developers to integrate the model into their own applications, customize it for specific use cases, and contribute to its ongoing development. The unified action schema further facilitates the adaptation of the model to various interfaces.
With the release of Utars 1.5, Bite Dance has set a new standard in GUI automation and AI-driven workflows. Its ability to perceive, reason, and act in a cohesive manner positions it as a powerful tool for developers and businesses looking to enhance their digital interactions. The open-source nature of the project encourages innovation and collaboration, paving the way for future advancements in AI technology.
Q: What is Utars 1.5?
A: Utars 1.5 is a significant upgrade to Bite Dance's vision language agent that treats your screen as a single image, allowing it to read, reason, and manipulate content directly.
Q: What are the model sizes available in Utars 1.5?
A: Utars 1.5 features three model sizes: a lightweight 2 billion parameter model, a mid-range 7 billion model, and a powerful 72 billion variant.
Q: How does Utars 1.5 perceive graphical user interfaces?
A: Utars 1.5 introduces a sophisticated perception system that analyzes various graphical user interfaces, synthesizing multiple types of perception data to recognize elements and their functions accurately.
Q: What is the reasoning capability of Utars 1.5?
A: Utars 1.5 can distinguish between intuitive and deliberate thinking processes, allowing it to break down tasks and learn from errors through an inner monologue before executing actions.
Q: How does Utars 1.5 perform in benchmarks?
A: In performance benchmarks, Utars 1.5 achieved a 42.5% success rate in complex tasks, outperforming previous models and showcasing its versatility in handling multi-step tasks.
Q: Is Utars 1.5 open-source?
A: Yes, Bite Dance has made Utars 1.5 accessible to the community by releasing its model weights and training scripts under an open-source license.
Q: What is the significance of Utars 1.5 for GUI automation?
A: Utars 1.5 sets a new standard in GUI automation and AI-driven workflows, enabling developers and businesses to enhance their digital interactions through its cohesive perception, reasoning, and action capabilities.