Claude’s thinking summaries are off by default now

A Reddit teardown puts the cost of Claude’s thinking summaries at 0.06% to 0.12% of Anthropic’s $47 billion in annual revenue. The bigger number is on the compute side: the summarization pass adds roughly 33% to decode load on every request with thinking, and decode is the part that hits a ceiling when capacity is tight.
In Claude Code, the beta header redact-thinking-2026-02-12 pulled thinking out of the terminal display, and version 2.1.90 on April 1 stopped generating summaries in interactive sessions entirely. You can get them back with showThinkingSummaries: true in settings.json. On claude.ai, the display disappeared around July 24.
The author is clear that tying the shutoff to a capacity crunch is his own inference. Fable 5 landed in Max and Team Premium on July 20 with half the usual limits, and the claude.ai change followed about four days later.
Up next
Amodei: distillation could cut China’s lag to a few months
Dario Amodei says industrial distillation could cut China’s lag behind the US frontier to a few months, but won’t produce capabilities that match or beat American models. His fix: target the distillation operations, not open weights as a category.
Related stories
- Anthropic introduces J-Space for hidden reasoning
- No multiplayer in Claude Code Projects yet, Anthropic says
- Claude Code on Pro lists Haiku 5.5 as the recommended default
- Claude Code on Opus 5.5 refuses to clone existing games
- Your claude.ai skills follow you into Claude Code 2.1.275
- Claude Code projects can burn your limits faster
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

