Significant intelligence regression since 200K→1M context window upgrade

User Feedback

Since the context window was increased from 200K to 1M, there has been a very noticeable drop in intelligence and comprehension quality. This is making it nearly impossible to work with Claude Code effectively.

Specific issues observed:

1. Repeated failure to understand user intent — User has to repeat themselves multiple times. Claude misinterprets clear instructions and goes in wrong directions.

1. Excessive unnecessary questions — Instead of understanding and acting, Claude presents multiple options and asks "which one?" when the answer is obvious from context.

1. "Did I understand correctly?" loops — Claude keeps asking for confirmation instead of just doing the work. This creates frustrating back-and-forth cycles.

1. Acting without understanding — Claude makes changes without properly understanding what was asked, then has to undo and redo work multiple times.

1. GPT-like behavior — The responses feel more like generic pattern matching than genuine comprehension. The user specifically noted: "I was already upset comparing you to GPT, but now you're even worse than GPT."

Context

• Model: Claude Opus 4.6 (1M context)
• Platform: macOS, Claude Code CLI
• Task: UI/UX design optimizations for a CustomTkinter app
• The user is an experienced developer who gives clear, direct instructions in Turkish

Impact

The user is considering switching to GPT due to these quality issues. This regression has been ongoing for days and is severely impacting productivity.

Expected behavior

Claude should:

• Listen carefully and understand instructions the first time
• Not ask unnecessary clarifying questions when intent is clear
• Act on instructions without creating option menus
• Maintain the intelligence level from the 200K context window era