Model Drift Observatory

Composite drift score over time by model

draft view / illustrative data
Model drift scores over time A line chart showing illustrative composite drift z-scores for GPT flagship, Claude Sonnet, Gemini Pro, and DeepSeek control from launch through day 180. +2z +1z 0 -1z -2z Launch Day 45 Day 90 Day 135 Day 180 GPT flagship Claude Sonnet Gemini Pro DeepSeek control
GPT flagship Claude Sonnet Gemini Pro DeepSeek control