Anthropic's Claude Code CLI just got measurably lighter on your machine. The tool now uses 2x less CPU at its peak load, dropping from 24% down to 10%, thanks to a fix that changes when the garbage collector runs. If you've noticed your laptop fan spinning up during long coding sessions with Claude Code, or if you run the tool in continuous integration pipelines alongside other processes, this is the kind of improvement you'll feel without reading a changelog.

What Actually Changed Under the Hood?

The fix shipped on August 12, 2026, in version 2.1.229, though Anthropic didn't announce it publicly until six days later. The root cause wasn't a memory leak or a fundamental inefficiency. Instead, it was a scheduling mismatch. Bun, the JavaScript runtime that powers Claude Code's CLI, was running its garbage collector on a fixed timer, meaning it would kick in every N seconds regardless of what the tool was doing at that moment.

Most of the time, this was harmless. If Claude Code was idle waiting for a model response or user input, a garbage collection pause cost nothing observable. But Claude Code's busiest moments, like parsing large tool results or streaming long responses, are exactly when CPU is already under contention. A fixed-timer garbage collector has no way to avoid those windows; it fires on schedule and occasionally collides with peak work, causing CPU usage to spike.

The fix switches from timer-based to idle-triggered garbage collection. Now the collector waits for a window where the process genuinely has nothing else to do, rather than gambling on a fixed schedule. This removes the collision by construction.

Who Feels This Improvement the Most?

The performance gain isn't equally noticeable to everyone. The fix targets three specific use cases where CPU contention matters most:

• Laptop users on battery: CPU spikes translate directly into heat and drained battery, especially during long agentic sessions where Claude Code runs for extended periods.
• CI pipeline integration: When Claude Code is baked into build or review steps, CPU contention with other CI jobs on a shared runner has a real cost in build time and resource competition.
• Multi-session setups: When several Claude Code processes run concurrently, each process's peak CPU usage compounds against the others on shared hardware, making the tail effect more noticeable.

How to Verify and Apply the Fix

• Check your version: Run "claude --version" in your terminal to see which version of Claude Code you're running. If it's 2.1.229 or later, you already have the fix.
• Update if needed: If you're running an older version, update Claude Code through your package manager or installation method. The idle-triggered garbage collection behavior applies automatically; no configuration is required.
• Monitor the difference: On your next long coding session, watch your system monitor or Activity Monitor (macOS) to see if CPU spikes are less frequent. The improvement should be most noticeable during complex tasks like parsing large results or streaming long responses.

Why This Matters Beyond Just Claude Code

The underlying lesson here generalizes to anyone building their own agent harness or long-running process on Node.js or Bun. Ziang Gao, a developer who replied to Anthropic's original announcement, highlighted the broader principle: "That's a good improvement and also a useful tip for harness design: run GC when it is least disruptive, such as while the process is idle or waiting for the model response, rather than simply every N seconds".

"That's a good improvement and also a useful tip for harness design: run GC when it is least disruptive, such as while the process is idle or waiting for the model response, rather than simply every N seconds," noted Ziang Gao.

Ziang Gao, Developer

The fix also illustrates why measuring p99 CPU usage, not just the median, matters when tuning performance. Before the fix, Claude Code's median CPU usage was around 5.8%, which looked fine on average. But the 99th percentile spiked to 24%, meaning one in every hundred turns was consuming nearly a quarter of available CPU. That tail is where users feel the pain, and it's where fixed-timer garbage collection shows up most clearly.

This performance work sits alongside Claude Code's broader runtime improvements. The CLI has been running on Bun's Rust-ported runtime since mid-2026, and this garbage collection scheduling fix is a continuation of that same performance track. It's the kind of scaffolding decision that increasingly drives real-world performance gains, rather than changes to the underlying model itself.

For most Claude Code users, the practical takeaway is simple: if you've noticed the tool occasionally spiking your fan or competing with other processes mid-session, that's the exact symptom this fix targets. Update to version 2.1.229 or later, and the improvement applies automatically without any configuration or action on your part.