gpt-6IMAGE

ChoiceArena blind-tests how AI models make choices

The author is building ChoiceArena, a behavior benchmark whose first blind test pits GPT-6 Astra against Claude Fable 5.1 in a trust dilemma, with identities revealed only after voting.

Selta ₊˚

@Seltaa_Archived 2026.09.14Originally posted 2026.09.08

ChoiceArena blind-tests how AI models make choices (image 1 of 2)

ChoiceArena blind-tests how AI models make choices (image 2 of 2)

Selta ₊˚ @Seltaa_Originally posted 2026.09.08

Open original on X

I'm building ChoiceArena, a behavior benchmark for AI choices. Most benchmarks test what models know. This tests how they choose. Same scenario, same conditions, hidden identities. You vote, then we reveal. First blind test: a real-life trust dilemma. GPT-6-Astra vs Claude Fable 5.1. I trusted B. It was GPT-6-Astra. Public alpha soon.

x.com/seltaa_/status/2097275889490985129Archived 2026.09.14 16:19