Web-search leaderboard launches with tests of 24 AI models
The creators say they made their internal evaluation method public to compare search performance end to end, with cost and latency alongside accuracy.
TLDR
The team announced the leaderboard on September 15, 2026, saying it aims to help people choose a model for search-heavy work. It describes tests of 24 frontier models using the same search, with and without web access. The team says its previously internal methodology measures accuracy alongside cost and latency.
Combined views
9.2K
1 Source, first seen 15d ago