Research
Benchmarks
Numbers the papers report on named benchmarks, with the conditions that make them comparable or not. Every figure is self-reported by the paper claiming it.
- 3 papers
BrowseComp
score
- 3 papers
ALFWorld
success rate
- 2 papers
GDPval-AA v2
Elo
- 2 papers
HR-Bench 4K
accuracy gain
- 2 papers
Hang Seng Index
net Sharpe ratio
- 2 papers
LIBERO
success rate
- 2 papers
LongVideoBench
accuracy
- 2 papers
Nasdaq-100
net Sharpe ratio
- 2 papers
OSWorld-Verified
score
- 2 papers
RoboTwin 2.0
success rate
- 2 papers
Terminal-Bench 2.0
Pass@1
- 2 papers
Terminal-Bench 2.1
accuracy
- 2 papers
WebShop
success rate
- 2 papers
ZoomBench
accuracy gain
- 2 papers
human evaluation
overall score
- 2 papers
AgentDyn
ASR@10
Spotted a wrong number? Tell us and it will be checked.