Claude keeps predicting long execution times which are wrong
retained text
Is anyone else seeing this -- I asked Claude Chat and Cowork, in Opus 5.5, to plan a task and Cowork said it will take at least 2 hours. I pushed back and said to him "Your previous execution times were very reasonable, minutes not hours". He came back and said, "You are right, we should be able to do this in a lot less time". Then he proceeded to execute the task, which also involved checking a few websites on Chrome via Claude on Chrome, and he was finished in about 30 mins. Is this basically continuing hallucinations about what they have been trained on -- which is that they have been trained on what it takes in amount of time for humans to execute?
document fetched 2026-11-15 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xw9gzu/claude_keeps_predicting_long_execution_times