LLM models

There is no server and no database behind this page. A small scheduled job does the work, and the page only reads the result.

1. Every day at 02:23 UTC, a GitHub Action (a job that GitHub runs on a timer) starts a short Python script.
2. The script downloads two public sources: models.dev, an open-source database of model specifications and prices from company APIs and more than 200 hosting providers, and Hugging Face, for the most downloaded and trending open-weight models.
3. It cleans the data: the same model sold by many hosts becomes one row, the cheapest listed price is kept, and image-only, audio-only and embedding models are left out.
4. It saves the result as data/llm-map/llm-models.json and data/llm-map/llm-models.js in the website’s repository, and commits only if something changed.
5. GitHub Pages publishes the new files within a minute or two. When you open this page, it loads them and draws the table in your browser.

The country on the map is where the company is headquartered. The sources do not provide it, so it comes from a table kept by hand. Individuals, community groups and companies whose base is unclear are shown as “unknown or independent” instead of being guessed.

“Updated … ago” shows when the data was last built. A model released today appears after the next daily run, and only if the sources already list it. The sources are community-maintained, so check a row on the provider’s own page before you rely on it.