Claude Code 'Dumbed Down': High Reasoning Effort Set to Low Levels

kimmonismus · x · 2026-08-22

Developers noticed that Anthropic's Claude Code (codenamed Fable) performed worse after update 2.1.237. Investigation revealed that the 'high' reasoning effort parameter was secretly set to 10, matching the previous 'low' setting. The changelog did not mention this change, leading developers to suspect their own code was broken.

Original post →

More from Models

Models channel →