Nerf Watch - AI Shrinkflation Index

We track each proprietary AI model's competitive ranking among all proprietary coding models available via API, measured weekly. When a model drops from #2 to #6, it means 4 models now score higher than it did - whether because competitors improved or the model itself changed. This captures the real competitive position that determines what users should pick.

Ranking drops happen for several reasons: new competitors launch with better specs (context window, capabilities, output length), existing competitors get updates, or a model's own specs change. Between w08 and w14, the proprietary coding model pool grew from 108 to 156 models. Even a model with unchanged quality will drop in rank as better alternatives enter the market.

Open-source models have published, immutable weights. Once released, they cannot be silently changed by a provider. Any ranking shifts for open-source models come from new competitors entering, not from the model itself changing. The shrinkflation concept - paying the same price for a relatively worse product - only applies to proprietary API models.

Shrinkflation Score (0-100) combines rank decline and price changes: each position dropped scores 3 points, and price increases add 2 points per percent increase. Formula: clamp(rankDelta * 3 + max(0, pricePctChange * 2), 0, 100). A model that dropped 10 positions with unchanged pricing scores 30. Higher = worse for users.