Avenir-Web: Human-Experience-Imitating Multimodal Web Agents with Mixture of Grounding Experts

TL;DR

Avenir-Web is a novel multimodal web agent that uses a mixture of grounding experts and experience-imitation planning to improve long-horizon task execution on complex, dynamic web interfaces, achieving state-of-the-art open-source performance.

Contribution

It introduces a new web agent architecture combining grounding experts, procedural priors, and adaptive memory, advancing open-source capabilities for complex web task automation.

Findings

Surpasses prior open-source agents on Online-Mind2Web benchmark

Achieves performance parity with top proprietary models

Demonstrates robustness across diverse web interfaces

Abstract

Despite advances in multimodal large language models, autonomous web agents still struggle to reliably execute long-horizon tasks on complex and dynamic web interfaces. Existing agents often suffer from inaccurate element grounding, the absence of site-specific procedural knowledge, and unstable long-term task tracking and memory, particularly when operating over complex Document Object Model structures. To address these limitations, we introduce Avenir-Web, a web agent that achieves a new open-source state of the art on the Online-Mind2Web benchmark in real-world deployment. Avenir-Web leverages a Mixture of Grounding Experts, Experience-Imitation Planning for incorporating procedural priors, and a task-tracking checklist combined with adaptive memory to enable robust and seamless interaction across diverse user interface paradigms. We evaluate Avenir-Web on Online-Mind2Web, a rigorous benchmark of live and user-centered web tasks. Our results demonstrate that Avenir-Web significantly surpasses prior open-source agents and attains performance parity with top-tier proprietary models, thereby establishing a new open-source state of the art for reliable web agents on live websites.