Researchers release WebAnything framework for automated web agent training
Researchers introduced WebAnything, a fully automated framework that recreates functional, reusable web environments from real URLs without manual setup. The system uses an Explorer-Builder to capture browser-observable structures and interactions, while an independent verifier ensures accuracy and guides repairs if needed. It supports 244 website artifacts across 14 domains and 53 categories,…
Key points
- WebAnything automates web environment reconstruction from real URLs without manual per-site engineering
- Supports 244 website artifacts across 14 domains and 53 categories, enabling 12,984 executed tasks
- Fine-tuning Qwen3.5-9B on 5,856 trajectories boosted task success rates by up to 13.33%
Fine-tuning Qwen3.5-9B on 5,856 interaction trajectories improved task success rates by 9.9% (REAL), 6.25% (Odysseys-Easy Rubric Avg), 6.32% (Task Avg), and 13.33% (WebArena-Verified Overall). The code is available on GitHub.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- AI startups aim to automate formal software verification with $365M funding · 2 src
- Researchers find instruction exposure unlocks rhetorical patterns in LLMs · 1 src
- New guide introduces LLMs for scientists and professionals · 1 src
- Researchers introduce Seq@K to test language models with sequential feedback · 6 src
- Researchers propose rank-aware method for private listwise LLM alignment · 1 src
Comments
via GitHub Discussions