Question: Model quality variation at different times of day?

Observation

User reported noticeable quality degradation when using Claude Code with Opus 4.5 at approximately 4am UTC (January 28, 2026). The model appeared to perform significantly worse compared to other times.

User's question

Is there any throttling, quantization, or use of smaller/different models during peak hours or specific times on Anthropic's servers?

Details

• Model: Claude Opus 4.5 (via Claude Code CLI)
• Time observed: ~4am UTC
• Perceived behavior: Described as "lobotomized" - noticeably worse reasoning and task completion
• Comparison: Same user noted significantly better performance at other times (same day, different hours)

Other reports

Another user reported similar observations on Hacker News:
https://news.ycombinator.com/item?id=46648900

Questions for Anthropic

1. Does Anthropic apply any form of throttling based on server load?
2. Are different model versions/sizes served at different times?
3. Is there any quantization applied during high-demand periods?
4. If none of the above, what could explain consistent quality variation at specific times?

This is a question/inquiry rather than a bug report. Understanding infrastructure behavior helps users plan their work and set appropriate expectations.

---

This issue was created with the help of Claude Code during daytime, when he's as smart as Lex Luthor 🧠