Bao Yu recommends /autocompact 300k to manage Claude Code token usage

What happened

Bao Yu recommends running `/autocompact 300k` in Claude Code, saying 300K was sufficient in his own use compared with the quoted recommendation of `/autocompact 400k`. Quoting @lydiahallie, he explains that auto-compact replaces the conversation with a short summary and triggers at about 967K on 1M models. Earlier compaction reduces the history included in subsequent requests; cached reads are cheaper but still count toward usage.

Read next

• More changes in today’s briefingGoogle DeepMind announces Gemini 4 Carbon with cybersecurity and long-horizon capabilities

The story so far

Bao Yu recommended running `/autocompact 300k` in Claude Code to reduce the conversation history included in subsequent requests, saying 300K was sufficient in his own use compared with the quoted recommendation of `/autocompact 400k`. Quoting @lydiahallie, Bao Yu explained that auto-compaction replaces the conversation with a short summary and normally triggers at about 967K tokens on models with a 1M-token context window. His recommendation moves that compaction earlier. He also noted that cached reads are cheaper but still count toward usage.