BREAKING
Arena Adds Factuality to Rankings
How the Factuality Signal Works
1
Sample atomic claims
↓
2
Verify via search
↓
3
Compute truth rate
↓
4
Blend with preference
0
M
Text claims
0
K
Search claims
0
%
Factuality weight
True Claim Rates by Arena
Text true rate
87
Search true rate
89
Winners and Losers
Climbers
Gained
●
GPT-5.5 up 13 to No. 7
●
GPT-5.2-search 11th to 3rd
●
Xiaomi 9th to 6th
Fallers
Dropped
●
Muse Spark 7th to 20th
●
Meta 2nd to 5th by lab
●
Google models trended flat
Factuality Debate Begins
AI NEWS BLITZ
Arena just added factuality signals that are reshuffling how top AI models rank.