it's *SWE-bench* guys, please and thank you
1:09 PM 路 Jun 2, 2026 路 974 Views
SWE-bench co-creator Ofir Press shares data showing AI agents scale from 34% to 75% accuracy over 1,000 steps
Performance gains scale logarithmically over extended step sequences.
--0--
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most ActivityTimeline
VIEWS429BOOKMARKS1LIKES3REPLIES1
Luca Soldaini 馃巰@soldni
@OfirPress 馃ゴ letting the team know!
Ofir Press@OfirPress
it's *SWE-bench* guys, please and thank you
7hViews 429Likes 3Bookmarks 1