Skip to main content
Models: 10
Dimensions: 26
Trials: 56,640
Pre-registered: osf.io/et4nf

Model Evolution

AI models change. The signals one version rewards aren't always the signals the next version rewards. When that happens, content strategies built for the old model quietly stop working. These are the transitions worth knowing about.

Why this page exists

When OpenAI ships a new GPT, or Google updates Gemini, the model usually keeps the same name in marketing copy but its underlying behavior can shift materially. A brand that was getting recommended for “best mid-priced X” queries may suddenly stop appearing — not because the brand changed, but because the model did. We track those changes here, with concrete advice on what to update when.

GPT-5.4 → GPT-5.5

April 2026

The signals that worked on GPT-5.4 don't work the same way on GPT-5.5. Five signals flipped direction; specificity became a top driver; comparison framing collapsed.

r = −0.15
fingerprint correlation
Read the analysis →

GPT-4o → GPT-5.4 lineage

Coming soon

How OpenAI's behavioral fingerprint evolved across GPT-4o, GPT-5.2, GPT-5.3, and GPT-5.4 — eight months of training shifts.

Gemini Pro vs Flash

Coming soon

How model size affects behavioral fingerprint. Same family, different parameter counts, materially different recommendation patterns.

Want to know how your brand stacks up on the current models?

The AI Commerce Assessment runs your brand against all 11 models we track, including the most recent releases.

Get your assessment →