Free · No account · Runs on your machine
Is your Claude Code or Codex plan worth it?
One command adds up the last 30 days of your Claude Code and Codex usage at API prices and puts it next to what your plan costs. It reads the logs already on your machine and sends nothing.
curl -fsSL https://ganivra.com/check.sh | shRead the script before running it: check.sh downloads check.py and runs it with Python 3.
Use Claude Code? The free Claude Code plugin goes deeper inside Claude Code: what loads before your first prompt, MCP servers and skills you never use, and a cost status line.
Illustrative output
What you get in about two seconds.
Claude Code · last 30 days · 22 active days
Usage at API prices $1,842.17
claude-opus-5-5 $1,610.40 61.2M input · 2.1B cache reads · 4.8M output
claude-sonnet-5-5 $231.77 9.4M input · 310.2M cache reads · 1.1M output
Busiest 5 hours $96.10 from Tue 18 Nov 14:00
Your plan Claude Ultra, $200 a month
→ Your plan covered 9.2x its price in usage. Keep it.
Codex · last 30 days · 6 active days
Usage at API prices $41.08
gpt-6-astra $41.08 2.3M input · 18.0M cache reads · 410.2k output
Your plan ChatGPT Business seat, $25 a month
→ About break-even: $41.08 a month at API prices against $25.
Top projects: checkout-api $1,204.90 · mobile-app $310.15 · other $368.20
Nothing was sent anywhere.What it reads
- Token counts, model, time and folder name from
~/.claude/projectsand~/.codex/sessions - The public LLM price list Ganivra maintains
- Your plan, which it asks once and remembers in
~/.ganivra/check.json
What it never does
- Print, store or send prompts, responses or code
- Read keys, tokens or account files
- Send usage anywhere, unless you add
--import - Keep full paths: only a folder's name, such as
checkout-api
How the answer is worked out
A month of usage at API prices, against your plan's price.
| Usage ÷ plan price | What it says |
|---|---|
| Above 1.5 | Your plan covered that many times its price. Keep it. |
| 0.5 to 1.5 | About break-even. |
| Below 0.5 | Paying per token would have cost less. The plan also covers the chat app, so weigh that too. |
With fewer than 5 active days it shows the numbers without a verdict. If the logs cover less than the period, the figures are scaled to a month and it says so.
Options
Add them after sh -s --, for example curl -fsSL https://ganivra.com/check.sh | sh -s -- --claude-plan max.
For engineering managers
Running a team on Claude Code or Codex seats?
The check answers for one laptop, once. Ganivra answers for every developer, every month: spend by developer, team, project and client, and who uses the most, from the telemetry both tools already send. Prompts and code are never sent.
After signing up, each developer runs the setup, then ganivra check --import. Your dashboard opens with the last 30 days already in it, not empty.
curl -fsSL https://ganivra.com/install.sh | sh
ganivra check --importQuestions
Is Claude Max worth it?
It depends on how much you use Claude Code. The check adds up your last 30 days at Anthropic's API prices and compares that with what your plan costs. If your usage would cost more than one and a half times the plan price on the API, the plan is a good deal; under half, paying per token would have cost less.
What does it read, and what does it send?
It reads the logs Claude Code keeps in ~/.claude/projects and Codex keeps in ~/.codex/sessions, and uses only token counts, model names, times and folder names. It never prints, stores or sends prompt, response or code text. The only request it makes is for the public price list, unless you add --import.
How accurate is the cost?
It uses each model's published list price, including cache reads, 5-minute and 1-hour cache writes, fast mode and long-prompt rates, the same rules as Ganivra's dashboard. Each response is counted once even though Claude Code writes it to several log lines. Codex fast mode isn't recorded in its logs, so those calls are priced at the standard rate.
Why doesn't it tell me to move from Max to Pro?
Anthropic and OpenAI don't publish the token limits behind each plan; they reset every 5 hours and every week. Usage alone can't show whether a smaller plan would have been enough, so the check shows your busiest 5 hours as evidence and leaves that call to you.
Does it work on Windows?
Not yet. It runs on macOS and Linux with Python 3.8 or later, which Macs with the developer tools already have.
What about Cursor, Copilot or the Claude and ChatGPT apps?
Cursor and Copilot don't keep usage logs in a stable format, and chat apps don't store usage on your machine, so the check covers Claude Code and Codex only.
Comparing seat prices before you buy? Use the AI coding tools cost calculator.