API reference — the public REST API under /api/v1: models, providers, comparisons, price history, and recent changes. No API key, no authentication. MCP connector — the same catalogue exposed as tools an AI assistant can call, over the Model Context Protocol.

Introduction — AI API Pricing

Q: Overview

AI API Pricing tracks what AI providers actually charge for their APIs — tokens, images, credits, whatever the unit — and publishes it as a public, free catalogue. Prices are pulled from providers' own pricing pages and official sources, normalized into comparable display units (per million tokens, per image, per credit), and re-checked on a daily ingestion run. There is no sign-up, no account, and no paid tier. Every price on the site is public, and the same data is available as a REST API and an MCP connector for assistants and agents.

Q: What the catalogue tracks, per model

Prices, one row per direction/modality/tier/cache combination (input, output, cached reads, batch and priority pricing, context-length brackets) — never collapsed into a single number, because two models with the same headline price can differ once you add cached-read discounts or a priority tier. Confidence, one of verified, high, medium, low, unverified — how much the catalogue trusts a price, and which sources agree or dissent on it. Lifecycle, when a model was released, when a deprecation was announced, and when it retires. History, kept as the points where a price actually changed — prices are step functions, so a stable model is one point in its history, not one per day.

Overview

AI API Pricing tracks what AI providers actually charge for their APIs — tokens, images, credits, whatever the unit — and publishes it as a public, free catalogue. Prices are pulled from providers' own pricing pages and official sources, normalized into comparable display units (per million tokens, per image, per credit), and re-checked on a daily ingestion run.

There is no sign-up, no account, and no paid tier. Every price on the site is public, and the same data is available as a REST API and an MCP connector for assistants and agents.

What the catalogue tracks, per model

Prices, one row per direction/modality/tier/cache combination (input, output, cached reads, batch and priority pricing, context-length brackets) — never collapsed into a single number, because two models with the same headline price can differ once you add cached-read discounts or a priority tier.
Confidence, one of verified, high, medium, low, unverified — how much the catalogue trusts a price, and which sources agree or dissent on it.
Lifecycle, when a model was released, when a deprecation was announced, and when it retires.
History, kept as the points where a price actually changed — prices are step functions, so a stable model is one point in its history, not one per day.

How the data is produced

A daily ingestion run pulls each configured source, reconciles disagreements by source weight and authority, and publishes a new dataset version. Every response from the site, the API, and the MCP connector carries that version (x-dataset-version on the REST API), so two callers reading the same version are reading the same numbers — the one thing this project cannot get wrong is an assistant quoting a different price than the website.

See Data sources for how to check a source's health and licence, and MCP connector for how an AI assistant can query the same catalogue directly.

Two ways in

API reference — the public REST API under /api/v1: models, providers, comparisons, price history, and recent changes. No API key, no authentication.
MCP connector — the same catalogue exposed as tools an AI assistant can call, over the Model Context Protocol.