Summary
ClawGUI is an open-source framework that provides full-stack infrastructure for GUI agents—AI systems that operate applications through visual interfaces using taps, swipes, and keystrokes rather than programmatic APIs, enabling them to control long-tail software that CLI-based agents cannot reach. The framework addresses three gaps in a unified platform: ClawGUI-RL offers the first open-source RL infrastructure for GUI agent training; ClawGUI-Eval enforces standardized evaluation pipelines across 6 benchmarks and over 11 models; and ClawGUI-Agent deploys trained agents to Android, HarmonyOS, and iOS through more than 12 chat platforms. An end-to-end trained model, ClawGUI-2B, achieves 17.1% success rate on MobileWorld GUI-Only. The authors argue that progress in GUI agents is bottlenecked less by modeling capacity than by the lack of coherent infrastructure. Paper by Fei Tang, Zhiqiong Lu, Boxuan Zhang, Weiming Lu, Jun Xiao, Yueting Zhuang, and Yongliang Shen (arXiv:2604.11784).
ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents
Research areas: cs.LG, cs.AI, cs.CL, cs.CV
Authors: Fei Tang, Zhiqiong Lu, Boxuan Zhang, Weiming Lu, Jun Xiao, Yueting Zhuang, Yongliang Shen
Published: 2026-04-13
arXiv: 2604.11784
Abstract (translated)
GUI agents drive applications through their visual interfaces instead of programmatic APIs, interacting with arbitrary software via taps, swipes, and keystrokes, reaching a long tail of applications that CLI-based agents cannot. Yet progress in this area is bottlenecked less by modeling capacity than by the absence of a coherent full-stack infrastructure.
This paper proposes ClawGUI, an open-source framework that addresses three gaps within a single unified platform:
- ClawGUI-RL — the first open-source RL infrastructure for GUI agents;
- ClawGUI-Eval — enforces standardized evaluation pipelines across 6 benchmarks and 11+ models;
- ClawGUI-Agent — brings trained agents to Android, HarmonyOS, and iOS through 12+ chat platforms.
An end-to-end trained
ClawGUI-2B achieves a
17.1% success rate on MobileWorld GUI-Only.
---
*Auto-collected on 2026-04-15.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177618485