microsoft/GELab-Zero-4B-preview-Sico-Evolution

VISIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 30, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

microsoft/GELab-Zero-4B-preview-Sico-Evolution is a 4 billion parameter GUI agent model, fine-tuned (LoRA) from the GELab-Zero-4B-preview base model by Microsoft. Optimized for GUI automation, it was trained on Microsoft Edge and Copilot UI trajectories using an iterative model evolution pipeline. This model achieves a dominant 82.9% Task Success Rate, significantly outperforming its base model and exceeding several closed-source and open-source state-of-the-art models in GUI task performance.

Loading preview...

GELab-Zero-4B-preview-Sico-Evolution: A GUI Agent for Digital Workers

This model, developed by Microsoft as part of the Sico open-source platform, is a 4 billion parameter (4B) GUI agent. It is fine-tuned using LoRA from the GELab-Zero-4B-preview base model, specifically on UI trajectories from Microsoft Edge and Copilot.

Key Capabilities & Performance

The GELab-Zero-4B-preview-Sico-Evolution model leverages a general-purpose GUI model evolution pipeline, an iterative mechanism designed to continuously improve an agent's real task success rate across various GUI applications. This process has led to significant performance gains:

  • Dramatic Improvement: Achieves an 82.9% Task Success Rate, representing a substantial +43.1% absolute increase over the base model's 39.8% baseline.
  • Outperforms SOTA Proprietary Models: Demonstrates superior performance compared to leading closed-source models, including gpt-5.4 (79.7%), Claude-Opus-4.6 (81.3%), and claude-opus-4.7 (82.1%).
  • Exceeds Open-Source Competitors: Significantly surpasses other prominent open-source models like kimi-k2.6 (62.6%) and UI-Venus-1.5-30B (61.0%).

Good For

  • Building Digital Workers: Ideal for developers creating AI agents that interact with graphical user interfaces.
  • GUI Automation: Excels in automating tasks within applications like web browsers (Microsoft Edge) and AI assistants (Copilot).
  • Research in Agentic Evolution: Provides a strong foundation for exploring and developing co-evolving human-AI systems, as detailed in Microsoft's research on agentic evolution.

Finetunes
1 models