Odysseys benchmark shows frontier web agents achieve 44.5% success on 200 realistic long-horizon multi-site tasks using rubric evaluation, with efficiency at only 1.15%.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Odysseys: Benchmarking Web Agents on Realistic Long Horizon Tasks
Odysseys benchmark shows frontier web agents achieve 44.5% success on 200 realistic long-horizon multi-site tasks using rubric evaluation, with efficiency at only 1.15%.