The Real Performance History of AI Models: Exposing Post-Launch Changes

AI model performance doesn't remain constant after launch—labs often introduce performance degradation through quantization, aggressive content filtering, or behavioral changes to reduce compute costs. LM Arena's objective ELO leaderboard, built from thousands of crowdsourced human evaluations, provides transparent tracking of these hidden trends that marketing announcements typically obscure. This data reveals critical differences between API performance and consumer web interfaces, empowering users to make informed choices based on real-world model capabilities.