Claude Code: 37% CPU Reduction During Streaming and Lower Long-Session Memory Usage
Claude CodeClaude Code v2.1.191 delivers a 37% reduction in CPU usage during streaming responses by coalescing terminal text updates to 100-millisecond intervals rather than updating on every token. The same release also reduces memory growth in long coding sessions by trimming the terminal output cache. Together, these changes make Claude Code noticeably more efficient on resource-constrained machines and during extended multi-hour sessions.
Sources & Mentions
3 external resources covering this update
Under-the-Hood Performance Wins
Claude Code v2.1.191 ships two performance improvements that affect every user: lower CPU consumption during streaming and reduced memory growth in long sessions.
37% Less CPU During Streaming Responses
Every token Claude streams to the terminal previously triggered an immediate screen update. Version 2.1.191 switches to a coalescing strategy: text updates are batched and applied at most once every 100 milliseconds. The measured result is a 37% reduction in CPU usage during streaming responses. Laptops run cooler, fans spin up less, and battery life extends meaningfully for developers working without AC power. The 100ms coalescing interval is imperceptible to users; responses still feel instantaneous.
Reduced Memory Growth in Long Sessions
Claude Code's terminal output cache could grow unbounded over the course of a multi-hour session. Version 2.1.191 trims this cache, reducing the long-session memory growth observed when working on large codebases for extended periods.
MCP Reliability Also Improved
tools/list, prompts/list, and resources/list capability discovery requests now retry on transient network errors with short backoff. MCP OAuth flows also retry once after transient network errors, and headless environments skip the browser popup and go directly to the paste-the-URL prompt.