Guide
Why is Claude Code so expensive?
Claude Code cost is rarely the model being pricey. It is usually context you pay to move around.
Context is the bill
Every turn re-sends the working context. Without caching, a large project context is paid at full input price on each request. Plugins, MCP servers, and tool outputs all add tokens you may not see.
Cache writes add up
The first time context is cached, you pay a write — 1.25× input for 5 minutes or 2× for an hour. Reused reads are then cheap (0.1×), but writes you never reuse are pure overhead.
Measure it
JP the Cat breaks the cost down per model and shows the 5-minute vs 1-hour cache-write composition, so 'expensive' turns into a specific, fixable number.
Common questions
- Is a bigger model always more expensive?
- Per token, yes — but a well-cached session on a large model can cost less than an uncached one on a small model. Where the tokens go matters more than the model alone.
See your real numbers with JP the Cat
A light, local menu-bar meter for Claude Code, Codex, Grok, and Muse Spark — real cost, honest limits.
Download MeowPrefer Homebrew? Install the same signed, notarized app from JP's public tap.
brew tap 8bittts/jpthecat
brew trust 8bittts/jpthecat
brew install --cask jpthecat