Claude Code cost estimator
Rough daily/monthly projection from average tokens per session and sessions per day. Tune numbers against your local logs or the TokenCheat CLI.
For engineers and team leads deciding between API billing and a subscription, or budgeting agentic usage. The math: tokens per session × sessions per day × the per-token rate, projected to a 30-day month.
Model prices verified 2026-07-02 against provider pricing pages.
Monthly (30d) ≈ $158 — illustrative only.
How to read the results
The projection is only as good as your tokens-per-session input, and that number varies enormously with context setup: CLAUDE.md size, MCP tool definitions, and how often compaction fires. Measure it from your local session logs (or the TokenCheat CLI, which reads them) instead of guessing. Treat the result as an estimate and verify against your provider invoice.
FAQ
- How is the projection computed?
- Straight multiplication: average tokens per session × sessions per day × per-Mtok rate / 1,000,000, scaled to 30 days. Worked example: at $3/M input, a session that consumes 1M input tokens costs $3.00 on the input lane alone.
- Why might this differ from my actual bill?
- Subscription plans don't bill per token, cache reads are discounted (Anthropic prices them at roughly 10% of input per their pricing page), and real usage mixes models with different rates. The estimator uses one blended rate — a simplification by design.
- What's the biggest lever on tokens per session?
- Standing context that ships with every request. In our analysis of 100 public configs, the median CLAUDE.md was 964 tokens across 92 lines — files far above that, plus heavy MCP loadouts, inflate every single request.
- When should I use this vs the full audit?
- Use this for a quick budget number. The full stack audit inspects the setup behind the number — CLAUDE.md, MCP overhead, caching — and tells you which parts of your per-session cost are avoidable.
For your full setup, run the free stack audit — or see the 100-configs report.