About TokenSave

TokenSave is a set of free calculators for people who build with AI. Every tool runs in your browser, needs no sign-up, and is available in 41 languages. Alongside the tools we publish our own measurements of what AI actually costs to use.

Tools

• AI Token Counter — count tokens, characters and API cost for GPT, Claude and Gemini models, and see how efficiently your language tokenizes.
• AI Video Cost Calculator — compare what the same clip costs on Veo, Kling, Runway, Luma, Grok, FLUX, Seedance and more.
• AI Image Cost Calculator — compare the price per image of Nano Banana, GPT Image, FLUX, Grok Imagine, Seedream and Runway.
• Subscription vs API — see whether ChatGPT, Claude or Gemini plans or pay-as-you-go API pricing is cheaper for the way you chat.
• AI Coding Agent Cost — estimate what Claude Code, Codex or Gemini CLI would cost on the API per task and per month.

The blog explains the ideas behind the numbers: what tokens are, why some languages cost more, and how to keep AI bills down.

Our own measurements

Most numbers about AI token costs online are copied from somewhere else. We measure them instead, and publish the raw data so anyone can check it. Each test is written up on a dataset page with the tool versions, the date, the exact prompts, how many runs were made and what the test cannot tell you. The files are released under CC BY 4.0 on Zenodo (each with a DOI), Hugging Face, Kaggle and GitHub.

Some of what we have measured so far:

• The same prompt in 41 languages on Claude, Gemini and GPT. In English Claude 5.5 used 38% more tokens than GPT; in Korean, 82% more. The token counter uses these figures.
• Claude Code vs Codex on the same coding task, including how many tokens each spends before you type anything.
• Claude effort levels, CLAUDE.md size, long chats vs new chats and what a coffee break costs once the prompt cache expires.

When a number on this site is an estimate rather than a measurement, the page says so.

In October 2026 the editors of Let's Data Science re-checked one of our datasets against its raw logs and published an interview about what it shows and what it can't.

Where the prices come from

Prices are copied by hand from each provider's official API pricing page and the date they were checked is shown on every tool. When a model has no public per-unit API price, we show the rate of a public reseller API (Runway) and label it clearly. Prices change often, so always confirm on the provider's page before a large job.

Your text stays with you

Token counting happens entirely in your browser. What you type is never uploaded or stored. See the privacy policy for details.

Contact

TokenSave is built and written by Jonhisking, an independent developer in South Korea who started making games with AI without a coding background and built the token counter to understand his own usage limits. Found a wrong price, a wrong number or a missing model? Email contact@jonhisking.com or use the contact page.