AnalysisAI ModelsOctober 26, 2026

Reddit test: Opus 5.5 writes slop code but still passes checkpoints

Read original source →reddit.com

A Reddit user built a SlopCodeBench-inspired test where an agent adds features on top of its own old code at each checkpoint without breaking anything. Opus 5.5 produced sloppy code yet succeeded on the checkpoints.

1 source

More stories today

Open the live feed