WebArena
WebArena
Why it matters
In an evaluation on real, working websites, a GPT-4 agent scored 14.41% versus 78.24% for humans, nailing down the gap between demos and reality
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2307.13854Added in Week 35, 2026 · See that week