Microsoft GELab-Zero-4B GUI agent hits 82.9% task success on browser automation
Microsoft released GELab-Zero-4B-preview-Sico-Evolution, a 4-billion-parameter GUI agent that executes browser and application tasks from screenshots and text commands, trained on Edge and Copilot interaction data.
Microsoft's GELab-Zero-4B-preview-Sico-Evolution is a 4-billion-parameter GUI agent that automates browser and application tasks without manual clicks. The model interprets screenshots and text commands to navigate visual interfaces, trained on interaction data from Edge and Copilot. It lifts task success rates from 39.8 percent to 82.9 percent on Microsoft's internal GUI benchmark, outperforming GPT-5.4 and Claude Opus 4.7 on the same test set.
The model uses LoRA adaptation to specialize for interface control, fine-tuning a small subset of weights rather than retraining the entire base. Released in GGUF quantized format, it targets local execution on consumer GPUs via llama.cpp and similar inference engines. At 4 billion parameters, it is the smallest open-weight GUI agent to exceed 80 percent success on complex interface tasks.
