Tag
WebAnything presents a fully automated framework that reconstructs source-grounded, reusable web agent environments from real URLs, building 244 website artifacts supporting 12,984 tasks and showing that fine-tuning Qwen3.5-9B on its interaction trajectories improves web agent success rates across multiple benchmarks.
WebAnything 是一个全自动框架,能从真实 URL 重建可复用、源绑定的 Web 环境,用于可扩展的智能体训练与评估。在 244 个网站环境上微调 Qwen3.5-9B,在 REAL、Odysseys 与 WebArena-Verified 基准上取得显著成功率提升。
提出一种可扩展的弱到强验证器加固循环,通过在隔离沙盒中迭代发现并修复漏洞,自动防止智能体环境的奖励作弊,并在 Terminal Bench、KernelBench 和 SETA 上大幅降低攻击成功率。
RecreationWorld introduces a scalable and verifiable framework for hybrid computer-use agents, providing environments across five platforms and a benchmark for evaluation.
Introduces GUI-CC, a benchmark for evaluating the contextual consistency of GUI world models when used as environments for GUI agents, through offline and online interaction tracks.
Mercor announces joining the OpenEnv committee alongside Meta, PyTorch, NVIDIA, PrimeIntellect, and Hugging Face to guide the open foundation for agentic environments.
PatchWorld introduces a gradient-free framework that transforms offline trajectories into executable Python world models via counterexample-guided code repair, enabling interpretable and inspectable belief-state programs for planning in partially observable environments.
SkillEvolBench is a diagnostic benchmark for evaluating whether large language model agents can distill episodic experience into reusable procedural skills. It includes 180 tasks across six environments and finds that current agents often struggle to form robust reusable skills, with raw trajectory reuse often outperforming distilled skills.