Jonhisking
Jonhisking is the independent developer who builds and writes TokenSave, based in Seosan, South Korea. Every article on this site with the byline "Jonhisking" was written and checked by him, and every measured number comes from tests run on his own accounts and computer.
Background
He had no coding background. He started by building Android games with AI, describing what he wanted to Claude in plain language instead of writing code (vibe coding). The first one, Sperm Race, is on Google Play; the first devlog covers how it was made and the second devlog covers the 14-day closed test.
While building games he kept hitting the usage limits of his AI plans (free, then Pro, then Max 5x), and moving up a plan did not fix it. Trying to understand why led him to tokens, and to how much more some languages cost than English. TokenSave started as a token counter he built for himself and grew from there.
How the tests are done
- Tests run on real accounts and real tools (Claude Code, Codex, the providers' token-counting APIs), not on numbers copied from other sites.
- Each dataset page states the tool versions, the date, the prompts and how many runs were made, plus the limits of the test.
- Raw logs and CSV files are published under CC BY 4.0 on Zenodo, Hugging Face, Kaggle and GitHub, so anyone can check or reuse them.
- Prices are taken from each provider's official pricing page, and the date checked is shown on the page.
- When a number is an estimate rather than a measurement, the page says so.
Published datasets
- Token cost by language: 41 languages measured on GPT tokenizers · DOI 10.5281/zenodo.23427148
- Token cost by language on Claude, Gemini and GPT: 41 languages measured · DOI 10.5281/zenodo.23428350
- Token cost by data format: JSON vs YAML vs CSV vs TOON on Claude, Gemini and GPT · DOI 10.5281/zenodo.23435788
- Image input token cost by resolution: Claude vs Gemini vs GPT · DOI 10.5281/zenodo.23443906
- Token cost by programming language: the same programs in 10 languages on Claude, Gemini and GPT · DOI 10.5281/zenodo.23446563
- PDF token cost per page: Claude vs Gemini, and PDF vs pasted text · DOI 10.5281/zenodo.23446565
- Claude effort levels: tokens, time and cost for the same coding task · DOI 10.5281/zenodo.23446567
- Claude Code vs Codex: startup tokens, MCP tools and the same coding task · DOI 10.5281/zenodo.23448342
- Web search in Claude Code and Codex: tokens, searches and sources per question · DOI 10.5281/zenodo.23449842
- AI web search by language: what Claude Code and Codex search for, and cite, in 5 languages · DOI 10.5281/zenodo.23451506
- CLAUDE.md and AGENTS.md size: what instruction files cost in Claude Code and Codex · DOI 10.5281/zenodo.23453217
- Coding in your own language: does the prompt language change what Claude Code and Codex cost? · DOI 10.5281/zenodo.23454308
- What the Claude Code harness costs: plugins, subagents, plan mode, skills and hooks on the same task · DOI 10.5281/zenodo.23465633
- TokenSave Language Tax Index: extra tokens per language on Claude, Gemini and GPT
- Vibe coding with Claude Haiku, Sonnet and Opus 5.5: cost, time and whether the game works · DOI 10.5281/zenodo.23489448
- Long chat or new chat? What 30 requests cost in one Claude Code session, with /compact, and in fresh sessions · DOI 10.5281/zenodo.23489464
- Does being polite to AI cost more? Plain, polite and very polite prompts in Korean, Japanese and English · DOI 10.5281/zenodo.23489481
- The coffee-break tax: what your next Claude Code message costs after a 0 to 65 minute pause · DOI 10.5281/zenodo.23489501
All datasets are listed on the data page.
Elsewhere
- Hugging Face: huggingface.co/JOHNISKING
- Kaggle: kaggle.com/johnisking
- GitHub: github.com/johnisking
- X: @tokensaveapp
Contact
Found a mistake in a number or a price? Email contact@jonhisking.com or use the contact page. Corrections are made on the page and noted in its date line.