AI

WebAnything: From the Wild Web to Scalable Agent Environments

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/10/2026, 06:48:24 UTC.

Briefings

WebAnything is a fully automated framework that reconstructs source-grounded, reusable Web environments from real URLs without manual repair. It constructs 244 website artifacts across 14 domains and 53 categories, supporting 12,984 executed tasks. The system outperforms human-assisted WebHarbor on three fidelity metrics, achieving 12 wins, two ties, and no losses in a blind audit. Fine-tuning Qwen3.5-9B on 5,856 interaction trajectories improved REAL success by 9.9 percentage points, Odysseys-Easy Rubric Avg by 6.25 percentage points, and WebArena-Verified Overall success by 13.33 percentage points under the fixed harness.

Key points

• WebAnything constructs 244 website artifacts across 14 application domains and 53 categories.
• The framework supports 12,984 executed and accepted tasks compared to human-assisted WebHarbor.

Context

This research addresses the need for diverse, executable, and resettable website environments for reproducible training and evaluation of web agents, addressing the limitations of human-assisted mirrors and generative construction methods.

Limitations

Evidence is limited to the 244 artifacts constructed and 12,984 tasks supported; no additional background on the impact of these findings is supplied.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/10/2026, 09:57:26 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeibr6kosvzp4dzvrzzfiprpekbf6nfcyy6gl4wd75aicwcy62u33ye