HyperBrowseComp: A Multilingual and Multimodal Stress Test for Web-Browsing Agents

What happened: Researchers introduced HyperBrowseComp, a multilingual and multimodal benchmark for web-browsing agents, with 423 manually authored and human-validated questions across 13 languages. Key numbers: 423 questions, 13 languages. Why it matters: The benchmark targets persistent information seeking across languages and evidence modalities, with difficulty from discovering and connecting evidence on the open web. Models were evaluated using provider-native search and a shared external retrieval harness under a common agent protocol, alongside a human evaluation.

Researchers introduced HyperBrowseComp, a multilingual and multimodal benchmark for web-browsing agents, with 423 manually authored and human-validated questions across 13 languages.

The benchmark targets persistent information seeking across languages and evidence modalities, with difficulty from discovering and connecting evidence on the open web. Models were evaluated using provider-native search and a shared external retrieval harness under a common agent protocol, alongside a human evaluation.