I've been measuring 34 LLM APIs every day since August to see when they quietly change. One did.

Since August 20, a developer has been measuring 34 LLM APIs daily from 15 labs to detect unannounced changes. This involves running consistent private probes and comparing each model against its own past performance, without external rankings or LLM-based judgments. The results are graded by code, and each reading is hashed into a public transparency log to establish an undeniable baseline. So far, one model has shown a quiet change, confirming the developer's hypothesis about unannounced model alterations.

This report uniquely details a daily, automated measurement of 34 LLM APIs since August, providing concrete evidence of unannounced model changes unlike anecdotal observations.