Claude Code sent 33,000 tokens before reading the prompt
In a small comparison, Claude Code sent about 33,000 tokens before processing the user's prompt, while sent about 7,000. The ers logged every request between each coding tool and Anthropic's server, including the returned usage records. The test began after Claude Code appeared to consume the usage allowance much faster than .
The measurements indicated that Claude Code used far more tokens for its built-in setup and handled caching less efficiently. The study included one stated caveat, so the size of the gap should not be assumed to be identical in every setup.
Key points
- Claude Code's measured usage before the prompt was about 33,000 tokens.
- 's measured usage before the prompt was about 7,000 tokens.
- The test captured requests and returned usage records at Anthropic's .
- Claude Code showed higher built-in token use and less efficient caching.
- Running the same task in both tools can reveal hidden fixed costs.