prithivMLmods/Gliese-CUA-Tool-Call-8B

VISIONPricing:Input $0.4 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:32kPublished:Dec 14, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Gliese-CUA-Tool-Call-8B by prithivMLmods is a 7 billion parameter multimodal Computer Use Agent (CUA) model, based on Qwen2.5-VL-7B-Instruct, designed for GUI understanding and interaction. It specializes in visual grounding, UI localization, and generating structured JSON tool calls for actions like clicks and scrolls across web, desktop, and mobile environments. This model is optimized for agentic tool calling, enabling reliable interaction with real-world software interfaces and multi-step task automation.

Loading preview...

Gliese-CUA-Tool-Call-8B: A Multimodal Computer Use Agent

Gliese-CUA-Tool-Call-8B, developed by prithivMLmods, is a 7 billion parameter multimodal model built upon Qwen2.5-VL-7B-Instruct. It functions as a Computer Use Agent (CUA), excelling in understanding graphical user interfaces (GUIs), localizing UI elements, and executing actions across web, desktop, and mobile platforms. The model's core strength lies in its ability to produce structured JSON tool calls, facilitating precise and deterministic interactions for agentic pipelines.

Key Capabilities

  • GUI Localization and Visual Grounding: Accurately identifies UI elements (buttons, text fields, icons) across diverse layouts and resolutions.
  • Parsed Action Prediction: Predicts low-level UI actions (clicks, types, scrolls) with explicit coordinates, allowing for transparent grounding and debugging.
  • Structured Tool Calling: Outputs actions as structured JSON within <tool_call> blocks, ideal for agentic tool-calling pipelines.
  • Action Planning and Execution: Translates natural language instructions into step-wise UI actions for multi-step workflows.
  • UI-Based Question Answering (VQA): Answers questions based on the current screen state, including element verification and workflow guidance.
  • Cross-Platform Computer Use: Operates consistently across web, desktop, and mobile interfaces.
  • Multi-Step Task Automation: Handles complex tasks like form filling, settings configuration, and dashboard navigation.

Good for

  • Agentic computer and GUI control via tool calling.
  • UI localization and element grounding with coordinates.
  • Generating structured actions for Robotic Process Automation (RPA) systems.
  • Automated form filling and workflow execution.
  • UI-based question answering and verification.
  • Developing web, desktop, and mobile agent frameworks.
  • Building accessibility and productivity assistants.

Finetunes
4 models