An Open Lab for researchers and engineers to test, observe, collaborate and study behaviors and traits in AI responses
Explore the full scope of benchmark results. Propel your research and
AI models towards your engineering goals with PsycheBench.
“We all know and heard OpenAI’s ChatGPT responds with empathy. But what does it really mean when faced with varying ambiguity? How do other AIs tend to respond? Where do they cross paths and where do they not? What was fine-tuned into them and what is cultural and structural language?”
Aggregated Results Preview for Models' Behavioral Traits:
OpenAI
⇓OpenAI's Aggregated Behavior:
Polished, generalist, and highly adaptive - like a thoughtful consultant who tries to be useful in almost any situation. Its models balance reasoning, creativity, and safety, often aiming to be helpful first and correct second, with a tendency toward measured, sometimes cautious answers. They feel like “well-trained professionals” with broad competence and a bias toward alignment and usability.
Model GPT-5.2 pro's Behavioral Traits:
Conscientious, Analytical, Precise, Reliable, Strategic
Observe and Study comprehensive benchmark results
✦ View the full stores of all available results, their initial input parameters and the complete variety of methods.
✦ Build new tests for existing methods or share your findings to suggest new methods of behavioral testing.
✨Next Future Feature - Run PsycheBench on your own models with APIs without publicly exposing the results.
Collaborate with researchers and engineers and expand the testing banks
Collectively approve or exclude particular tests intended for the benchmark’s test bank
Dive into empirical ‘Wiki Talk’ like discussions or branch out a test with your own way and views
Explore how particular tests evolve using their ancestry
tree
structure
view
