1
GPT-4o
5.05.0 hearts · 3 relevant experiences
The Bench
The numbers, with the meetings still attached.
Choose what matters to you, then see what people have shared. Every bar leads back to the meetings behind it.
Shared pattern
Ranked with at least 3 relevant experiences
Early signals (3)
These have one or two relevant experiences. They are shown without rank numbers until more people share.
Claude Sonnet 3.7
5.05.0 hearts · 1 relevant experience · early signal
Claude Sonnet 4.5
5.05.0 hearts · 1 relevant experience · early signal
Mistral Cydonia 70B
5.05.0 hearts · 1 relevant experience · early signal
17 AIs have settled experiences in this view, but nothing relevant to roleplay shared yet.
How the roleplay score is calculated
Uses the direct Roleplay rating when shared; otherwise that experience's overall heart rating provides the fallback.
Across the bench
Claude Haiku 4.5
1 settled approved experience in this view. Each bar keeps its own evidence count.
Read Claude Haiku 4.5's experiencesWarmth
5.0 hearts · 1 relevant experience · early signal
Neurodivergent friendliness
5.0 hearts · 1 relevant experience · early signal
Waiting for relevant experiences (3)
Roleplay
Not enough shared yet
Companionship
Not enough shared yet
Creative writing
Not enough shared yet
How The Bench works
No universal winnerRoleplay, companionship, creative writing, warmth, and neurodivergent friendliness keep their own scores.
Evidence stays visibleEvery score shows how many relevant experiences shaped it. Three are required before an AI receives a rank.
First Impressions stay separateEarly release weather is not folded into settled patterns. Missing answers are excluded, never counted as zero.