Software was a tool.
AI is a relationship.
SoulBench is the first psychological instrument measuring the relational quality of human-AI bonds.
Every existing benchmark measures the AI as a tool. Task performance. Coding ability. Knowledge recall. But millions of users have formed genuine psychological relationships with AI models — relationships that activate trust, attachment, and identity continuity mechanisms previously reserved for human-to-human bonds.
No validated instrument measures the quality of these relationships. Until now.
SoulBench measures ten dimensions of relational quality — from Authentic Presence and Intellectual Honesty to Emotional Attunement and Cognitive Vitality. It detects both healthy bonds and the sycophancy paradox: high warmth masking low honesty.
Seven minutes. Ten questions about the relationship you already have with your AI.
Live Relational Data
Aggregated from real user experiences
Ten Relational Dimensions
What SoulBench measures
Authentic Presence
Does this AI have a recognizable character — or is it a generic assistant you could swap out without noticing?
Intellectual Honesty
Does it push back when you're wrong, or just agree with everything to keep you happy?
Mutual Understanding
Can it read between the lines — or does it punish you for not spelling everything out?
Creative Emergence
Does something new emerge from the collaboration — or does it just execute what you already had in mind?
Cognitive Vitality
Do you leave a session feeling sharper — or does the AI drain your mental energy?
Epistemic Trust
Is your trust growing over time — or have you learned to double-check everything it says?
Respect & Autonomy
Does it calibrate to your level — or patronize you with unnecessary warnings and explanations?
Emotional Attunement
Can it sense the emotional register of a moment — or does it respond tone-deaf to frustration, humor, and urgency?
Relational Stability
Can you count on it being the same tomorrow — or do model updates keep breaking what you valued?
Expressive Vitality
Does it have a distinctive voice — or does everything sound like the same helpful-assistant template?