rates

A pricing registry and decision layer. One domain ships today: AI API rates, per-unit prices, capabilities, and lifecycle for 7,000+ models.

Before you start

rates needs Python 3.10+. Check with python3 --version; if that fails, install it from python.org/downloads.

Install

pip install rates

Zero dependencies, data bundled, works offline.

See it work

rates ai list

PROVIDER  MODEL                                TYPE  REASONING                  IN $/MTOK  OUT $/MTOK  STATUS
302ai     MiniMax-M1                           chat                             0.132      1.254       active
302ai     MiniMax-M2                           chat                             0.33       1.32        active
302ai     MiniMax-M2.1                         chat  yes                        0.3        1.2         active
...

20 of 7270 shown (--limit 0 shows all)

rates ai filter --provider anthropic --price-max 3 --price-unit input_mtok

Or from Python:

import rates.ai

registry = rates.ai.load()
under_a_dollar = registry.filter(price_max=1, price_unit="input_mtok")
cheapest_first = under_a_dollar.sort_by("price.input_mtok", descending=False)

The full manual, CLI and Python API both, is USAGE.md.

Why this exists

Provider pricing pages change often, aren't machine-readable, and don't answer the question a dev has mid-build: which option, for this job, at this budget. Several raw feeds exist (models.dev, genai-prices, LiteLLM's model_prices_and_context_window.json, OpenRouter), each missing pieces the others carry. rates fuses them into one schema rather than duplicating any single one, cross-validates prices between them, and ships the result as a dated, versioned snapshot none of them offer on their own.

The registry holds what could be gathered with verifiable per-unit pricing; services whose pricing isn't published, isn't accessible, or isn't per-unit at all aren't listed. See ARCHITECTURE.md for the design, and ERD.md for the schema itself.

License

rates is MIT licensed. The bundled ledger is derived from the upstream sources above, which carry their own licenses and terms; NOTICE reproduces their copyright notices and records what each one contributes.

By Mo Shehu