Numbers only, measured from my own local Claude Code session logs (2026-11-02 to 2026-11-08, 7-day daily average).
- Calls per day: 5,729
- Cache-read tokens per day: 619M
- Output tokens per day: 5.1M
Share of cost by slice:
| Slice | Share of cost |
|---|---|
| Base (system prompt + tool list) | 38.2% |
| Model replies | 27.1% |
| User input + attachments | 13.0% |
| Subagents | 9.7% |
| Shell command output | 8.6% |
| File reads | 2.8% |
| Search (grep/glob) | 0.4% |
| MCP + other tools | 0.2% |
| Web fetch | 0.0% |
Cost weights: input 1, cache write 1.25, cache read 0.1, output 5 (input token = 1).
Takeaway: the biggest slice is the fixed context that gets re-read on every call, so shorter sessions help more than trimming tool output.
Top comments (0)