Process
Methodology
Trust comes from process. Here is exactly how the numbers on this site are collected and kept fresh.
01
Data model
Every model is a single JSON file in the public repository (data/models/). A file contains the context window, capabilities and a list of pricing entries — one per provider — each with input/output prices per 1M tokens, an optional cached-input price, the date it was last checked and a link to the official source page.
02
Sources
Only first-party sources: official provider pricing pages and documentation. Aggregators and blog posts are never used as a price source. If a provider changes a page structure, the entry is flagged for manual review instead of guessing.
03
Update cadence
A scheduled GitHub Actions workflow runs the update bot weekly. The bot re-reads source pages, diffs them against the current JSON and opens a pull request with the changes. A human reviews and merges — nothing auto-publishes. Each entry shows its own updatedAt date, so staleness is visible per row, not per site.
04
Units and normalization
All prices are stored in USD per 1 million tokens. Tiered pricing (e.g. different rates above a context threshold) is stored at the base tier and noted on the model page. Batch/offline discounts are out of scope for now.
05
Report an error
Spotted a wrong or stale number? Open an issue in the GitHub repository (https://github.com/no-way-dev/no-way-dev) with the model, the wrong value and a link to the official page. Corrections are typically reviewed within a day. You can also open a pull request directly — the data files are plain JSON validated by zod schemas.