About Token Barometer

Who

I am Dejan Strbad, an engineer in Croatia. I work on edge ML, FPGA inference and industrial systems. I write on Substack and post as @dstrbad on X.

Why I built this

Every LLM price list I found showed today's number and nothing else, so there was no way to see when a price changed or by how much. A token is not a unit: the same text is a different number of tokens on Claude, GPT, Gemini or Llama, so per-token prices are not directly comparable. Most cloud-vs-local comparisons assume the GPU is busy all the time and ignore the idle hours you still pay for.

How it works

Token Barometer tracks list prices for LLM API offerings: a model as served by a host, in USD per 1M tokens. Prices are collected daily from vendor pricing pages, official APIs and third-party aggregators, normalised to a common schema, and kept as an append-only price history so every change is visible. It also includes a cloud-vs-local cost calculator for comparing a hosted API against running an open-weight model yourself.

Each price has a source and a confidence level. When sources disagree, the most authoritative one wins: official vendor pricing pages first, then the OpenRouter API, then community aggregators (LiteLLM, models.dev). A large unexplained price jump is parked for human review rather than published automatically. See the full methodology for sourcing, normalisation and the calculator's formulas.

Sources

Data licence

Data is CC BY 4.0. If you use it, link back:
Token Barometer by Dejan Strbad, tokenbarometer.com

See the API page for JSON and CSV access.

Code

The code is closed for now. The data, the API, the methodology and every calculator formula are public.

Contact

Wrong price or broken page: open an issue at https://github.com/dstrbad/tokenbarometer-issues/issues. Everything else: @dstrbad on X.