We maintain an open dataset of LLM API list prices (about 100 models) where every row links to the provider’s official pricing page and carries the date it was verified. Since July we also log every change, so here is what moved in the last three months.
Price cuts
Price increases
Churn: 22 new models in September and October, 16 deprecation and retirement notices.
Try it: the LLM API Pricing Calculator Space has a price table, a monthly cost estimate for your token volumes, and the change log. It loads the data live, so it stays current.
Use the data: current prices are in aicostbudget-ai/ai-api-pricing; the history (dated snapshots and change events) is in aicostbudget-ai/llm-api-price-history:
from datasets import load_dataset
changes = load_dataset("aicostbudget-ai/llm-api-price-history", "price_changes", split="train").to_pandas()
print(changes[changes["change_type"] == "price_update"])
In the snapshots, each change is labeled either as a provider price change (with its event) or as a correction on our side, so dataset fixes don’t show up as fake market moves.
How the prices are verified: prices are only accepted from the provider’s own pricing page, docs, API or announcement; third-party aggregators are never used as evidence. Each release captures the source passage, validates the schema and tiers, and is rehearsed against the website and the Hugging Face export before publishing. Unknown prices stay empty, never zero.
Feedback welcome: which providers or pricing modes (batch, caching, long context) should we cover next, and is a price you rely on missing or wrong?