Generalization of MTI Profiles to Frontier Models

Determine whether Model Temperament Index (MTI) temperament profiles derived from small open‑weight language models (1.7B–9B parameters) generalize to frontier models such as GPT‑4, Claude, and Gemini Pro.

Background

The study measures MTI profiles for 10 small, open‑weight models ranging from 1.7B to 9B parameters. While these results support axis independence and construct validity within this range, the evaluation does not include frontier API‑served models.

Because deployment decisions frequently involve frontier models, establishing whether the four‑axis temperament structure and observed distributions extend to GPT‑4, Claude, Gemini Pro, and similar systems is critical for external validity and practical applicability.

References

Whether MTI profiles generalize to frontier models (GPT-4, Claude, Gemini Pro) is unknown.

MTI: A Behavior-Based Temperament Profiling System for AI Agents  (2604.02145 - Jeong, 2 Apr 2026) in Subsection 6.1 (Limitations) — Model scope and sample size