Original post
Connor Shorten#1057
Connor Shorten#1057
Bo@bo_wangbo
finally i managed to catch up the BrowseComp-Plus train with PPLX 0.6B (single/multi-vec) models:
6:30 AM · May 27, 2026 · 13K Views
Perplexity's Bo Wang says PPLX 0.6B late-interaction embedding model achieves 53% accuracy on BrowseComp-Plus benchmark
A GPT-5 and pplx-late pipeline reached 85.78% accuracy.
9196166924.7K
Sentiment
Sentiment building, check back later.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Posts from X
Most Activity
Most Activity
VIEWS6.7KBOOKMARKS31LIKES82RETWEETS7REPLIES4
Omar Khattab@lateinteraction
honestly loving this era of @perplexity_ai training and open-weight'ing colbert models
one nice side effect: you can compare pplx-late vs. pplx-emb to see the gains you get by just adding multi-vector interactions to a great model recipe
Bo@bo_wangbo
finally i managed to catch up the BrowseComp-Plus train with PPLX 0.6B (single/multi-vec) models:
27dViews 6.7KLikes 82Bookmarks 31
Antoine Chaffin@antoine_chaffin
Late interaction makes it look like it is easy 💁🏻♂️ It is a very cool results because it confirms the trend one more time, but also because the PPLX multi-vector model is multilingual!
Multilingual deep research is now available for everyone!
Bo@bo_wangbo
finally i managed to catch up the BrowseComp-Plus train with PPLX 0.6B (single/multi-vec) models:
28dViews 5.1KLikes 60Bookmarks 24
Strata@ChainZenit
@lateinteraction @perplexity_ai That multi-vector boost's been known since 2017 tho
27dViews 17