llm-price-tracker

llm-price-tracker fetches official published API token prices for major LLM providers, normalises them into one schema, and supports JSON snapshots and snapshot diffs.

Prices are collected from official provider pages only. Provider page structures can change without notice, so review fetched prices before using them for production billing or customer-facing estimates.

Supported providers

• OpenAI GPT API models: https://developers.openai.com/api/docs/pricing
• Anthropic Claude models: https://docs.anthropic.com/en/docs/about-claude/pricing
• Google Gemini models: https://ai.google.dev/gemini-api/docs/pricing
• Kimi / Moonshot AI models: https://platform.kimi.ai/docs/pricing/chat
• MiniMax models: https://platform.minimax.io/docs/guides/pricing-paygo
• Qwen / DashScope models: https://www.alibabacloud.com/help/en/model-studio/models

Installation

pip install llm-price-tracker

For local development:

python3 -m pip install -e ".[dev]"

Python 3.10 or newer is required.

Python usage

from llm_price_tracker import (
    diff_snapshots,
    fetch_all_prices,
    fetch_provider_prices,
    load_snapshot,
    save_snapshot,
)

prices = fetch_all_prices()
openai_prices = fetch_provider_prices("openai")
save_snapshot(prices, "prices.json")

old = load_snapshot("prices-old.json")
new = load_snapshot("prices.json")
diff = diff_snapshots(old, new)

CLI usage

llm-price-tracker list-providers
llm-price-tracker fetch --provider openai --output openai-prices.json
llm-price-tracker fetch --provider all --output prices.json
llm-price-tracker diff --old prices-old.json --new prices.json
llm-price-tracker diff --old prices-old.json --new prices.json --fail-on-change

The CLI exits non-zero when fetching, parsing, validation, or diff --fail-on-change checks fail. Use --ignore-errors with fetch --provider all to write successful providers while reporting skipped provider errors.

Output schema

Snapshots are JSON arrays of ModelPrice objects. Decimal prices are serialized as strings to avoid floating-point precision loss.

[
  {
    "provider": "openai",
    "model": "gpt-4.1",
    "input_per_1m": "2.00",
    "output_per_1m": "8.00",
    "cached_input_per_1m": "0.50",
    "cache_write_5m_per_1m": null,
    "cache_write_1h_per_1m": null,
    "cache_storage_per_1m_hour": null,
    "currency": "USD",
    "unit": "1M tokens",
    "source_url": "https://developers.openai.com/api/docs/pricing",
    "source_type": "html",
    "fetched_at": "2026-06-12T09:00:00Z",
    "modality": null,
    "billing_tier": null,
    "price_condition": null,
    "notes": null
  }
]

Some providers publish tiered or modality-specific prices. Those variants use optional modality, billing_tier, and price_condition fields instead of inventing new model names.

Provider adapters include lightweight required-model sanity checks for current official pages. Those checks are adapter arguments so downstream users can override them when providers rename, add, or deprecate models.

Development

python3 -m pip install -e ".[dev]"
pytest
ruff check .
black --check .

Normal tests use saved fixtures and do not call live provider pages. Live tests, when added, should be opt-in:

LLM_PRICE_TRACKER_LIVE_TESTS=1 pytest tests/live

Author

Eugene Evstafev hi@eugene.plus