Why My Claude-4-Sonnet Thinking Slowed to a Crawl in Cursor and How I Adapted

Picture this: I’m deep in a coding sprint using Cursor IDE when Claude-4-Sonnet’s “thinking” mode grinds to a crawl. We’re talking 0.3 tokens per second – like wading through molasses. And of course, it happened during my peak productivity hours. I had to find a fix, fast. After some trial and error, I figured out a few tricks that got me back on track.

The Frustrating Slowdown I Encountered

Out of nowhere, Claude’s thinking responses became painfully slow. Token generation plummeted to near zero. This wasn’t about my chat history or prompt length – even fresh conversations struggled.

What really stung? The non-thinking Claude still worked fine, but I needed that deeper analysis for complex coding tasks. The slowdowns seemed to target my busiest times: late mornings and afternoons. Waiting 15 minutes for a response? Not exactly productivity heaven.

What I Discovered About the Cause

After watching this pattern for days, I noticed something interesting. Slowdowns hit hardest between 10 AM and 2 PM CEST for me in Europe. Friends in Toronto and Istanbul reported similar midday slumps.

• Anthropic’s API load appeared to be the root cause – regional traffic spikes triggering throttling
• My account limits weren’t the issue (plenty of requests left), just high global demand
• Performance swung wildly minute-to-minute, pointing to server-side capacity limits

This wasn’t a Cursor bug – just Claude buckling under heavy traffic.

Steps I Took to Regain Control

I needed solutions that worked immediately. Here’s what actually helped:

• Switch to non-thinking mode instantly: For simpler tasks, I’d toggle off thinking mode to keep moving
• Reschedule my AI work: I saved heavy Claude thinking for early mornings or evenings when speeds were stable
• Keep backup models ready: When Claude lagged, I’d briefly switch to other Cursor-supported models
• Track request patterns: With privacy mode off, I monitored request IDs to spot recurring issues

Bonus benefit: avoiding charges for timed-out requests!

Key Takeaways for Smoother AI Work

This experience taught me a valuable lesson: sometimes the AI’s speed is out of your hands. My workflow now includes flexibility. I time my Claude sessions, keep alternatives handy, and never assume constant availability. While I’m hopeful Anthropic will boost capacity, these adjustments keep my Cursor workflow flowing – even when Claude decides to take a coffee break.