AI models shared 43% of their ideas on average, Arena reports
Arena says its analysis of 30,086 paired answers didn’t consistently support the expectation that models from the same lab or country would show greater conceptual overlap.
TLDR
Arena reports that models shared 43.1% of their ideas across 30,086 pairs of answers to the same user prompts. Some adjacent generations overlapped more: Grok 4.5 and Grok 4.6 shared 59.7%, while GPT 5.5 and GPT 5.6 shared 59.2%, according to Arena.
A higher token price didn’t necessarily mean a different perspective in the analysis. Arena says Fable 5 and DeepSeek V4 Pro shared 59.2% of their ideas despite a 12× difference in token price.
Combined views
89.1K
1 Source, first seen 19d ago