Developer says GPT quality tanked across the board: watered-down output, broken agent instruction following
KhanNdifor · x · 2026-10-30
Developer KhanNdifor vented that GPT quality has dropped across the board over the past month: GPT 5.5 used to write great short films, songs, and humor, but output now feels heavily watered down. Their NoSpoon agents see instruction following completely fall apart on gpt 5.6 and 6.1 sol, and they speculate compute constraints are to blame. Challenged with "maybe your taste evolved," the complaint echoes Kyrannio's side-by-side lyrics comparison — another data point in the ongoing silent-downgrade debate, though still lacking systematic benchmarks.
Related event: Users Report Across-the-Board Quality Decline in GPT Models(4 posts)→