WebHarbor suite provides real website environments for GUI agents
WebHarbor offers 153 executable dockerized environments with 3518 tasks for evolving GUI agent training and evaluation.
Stories per day · 30 days
2 RSS
Relatedreinforcement-learning4osworld3qwen3agent-training2android2benchmark2open-weights2adapters1agents1computer-use1computer-use-agents1grounding1huggingface1ios1jepa1llm-as-judge1mai-gym1mobile1
26 stories, newest first · one card per story
Stories tagged #gui-agents
Online RL framework trains UI agents on live iOS apps
An 8B planning-plus-grounding agent nearly septuples the SFT baseline on Android and ties Gemini-3.1-Pro on real iOS apps.
Adaptive Training Framework Cuts GUI Agent Rollouts by 56-61%
The two-stage method uses reusable checkpoints and action disagreement to improve RL efficiency on MobileGym and CUA-Gym-Hub.
GUI-JEPA scores mobile GUI actions in latent space without decoding
On AndroidControl it is 2.8–4.0x faster than a fine-tuned Qwen2.5-VL-7B agent while using 6% of the parameters.
OpenMobile-2 agents approach proprietary scores on phone benchmarks
The 27B model rose from 31.9 to 59.6 on real-device SPA-Bench, which the authors call the best open score.
IGG benchmark asks GUI agents to reveal hidden targets before clicking
Humans score 94.6%, GPT-5.5 79.4%, and the best open-source VLM only 31.6% on 601 tasks.
FocusGUI adapter cuts sequential distraction in omni-modal GUI agents
Across four frozen backbones, authors report Type Match up 33.2 points and Exact Match 3.0 times the frozen baseline.
OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
OpenGCUA open-sources a generalist computer-use stack that lifts OSWorld-v2 scores and cuts steps.
OS-Helix trains computer-use agents to mix GUI and code
OS-Helix-27B-RL reaches 66.44% on OSWorld-Verified, a further 42.9% relative gain after RL.
MAI-Gym: An Open All-in-One Platform for GUI Agent Evaluation and RL Training across Mobile and Computer Environments
MAI-Gym unifies GUI-agent evaluation and online RL; its 9B agent scores 78.4% on MobileWorld.
OpenMobile-2 open mobile agent trained on 110-plus apps
Authors report AndroidWorld 79.9, MobileWorld 50.4, and SPA-Bench rising from 31.9 to 59.6.
RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything uses world models to select actions and updates, improving code repair, GUI tasks, and lab planning.
LGWM verifies GUI agent transitions by predicting next-screen representations
On RSWT-BENCH, where screen-only detectors are at chance, the training-free score hits 0.987 AUC in 17 ms.
GUI-HARVEST: Self-Improving GUI Agents through Evidence-Driven Harness Evolution
GUI-HARVEST evolves GUI agent harnesses, lifting frozen Qwen3-VL by 12.33 points on OSWorld.
cua-speedrun: Standardized Benchmarking of the Speed of Computer-Use Agents
cua-speedrun standardizes virtual-machine infrastructure to measure computer-use agent speed, cost, and efficiency.
ComputerSD: Online Self-Distillation from Real-Time Feedback for Computer-Use Agents
ComputerSD turns real-time GUI feedback into online self-distillation signals for computer-use agents.
Relevance Does Not Imply Applicability: Experience Activation for Personal GUI Agents
ExpActivator activates only situation-applicable GUI history, lifting step success 28% while using far fewer tokens.
Jev-Mobile: Jev as an Executor for Mobile GUI Agents
Jev-Mobile splits VLM planning from fast GUI execution, cutting AndroidWorld API cost 73% at similar success.
OSWorld-Pro: Process-based Evaluation for Computer Use Agents
OSWorld-Pro scores computer-use agents on over 300 tasks; Claude Opus 5 reaches 75.7%.
Show HN: CUA-S1 – A System One Model for Computer Use
Cua open-sources CUA-S1, a fast 'System One' model for driving computer-use agents, launched on GitHub via Show HN.
RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents
RecreationWorld benchmarks hybrid computer-use agents that rediscover and reimplement apps across five platforms.
MintAct: A Unified Visual Agent for Digital Environments
MintAct vision-language models at 2B, 4B, and 8B unify UI grounding, multi-step navigation, and visual tool use, matching per-domain specialists.
Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agents
EvoSkill-GUI introduces training-free skill revision for GUI agents, boosting MobileWorld, AndroidWorld, and OSWorld success by up to 16.2%.
CADWorld: Computer-Use Benchmark for Long-Horizon Computer-Aided Design
CADWorld benchmark evaluates computer-use agents on 200 long-horizon FreeCAD tasks; best agent reaches 17.5% success versus 87% expert reference.