BROWSECOMP BENCHMARK FOR HARD-TO-FIND KNOWLEDGE

Summary from the TLDR AI newsletter

OpenAI's BrowseComp is a new benchmark of 1,266 problems designed to evaluate AI agents' browsing skills in gathering complex, hard-to-locate information online.