Paper
The peer-review-ready writeup of this study. We publish version-stamped PDFs as the dataset grows. The current paper covers 21 models (7 frontier + 14 historical) across 14 psychometric instruments with full token and cost accounting.
Available versions
Citation
@misc{personality-bench-2026,
title = {Personality Bench: A Cross-Lab Inventory of Frontier-LLM
Self-Presentation Across 14 Psychometric Instruments,
with Cross-Version Drift Analysis},
author = {Adams, Anthony David},
year = {2026},
note = {Published by EarthPilot.ai — Mission Support for
Spaceship Earth}
}Related
- Interactive drift charts → — same data as paper §4, with all family lineages overlaid
- Full cost ledger → — every billed API call with token counts
- Methodology notes →
- Source on GitHub →