Methodology: how we verify prices
What we measure, where the numbers come from, how often we check them, and where our estimates are weaker.
Sources
Every price on this site is copied from the official pricing page of the provider that sells the model:
- Anthropic: platform.claude.com pricing documentation
- OpenAI: developers.openai.com API pricing
- Google: aistudio.google.com pricing
We do not use third-party price lists, because they often lag behind or disagree with each other.
What we record
For each model we record the input price, output price, and cached-input price per million tokens in US dollars. We use the standard tier and, where a provider has a short-context and a long-context tier, the short-context price, noting the long-context price where it applies. We do not cover fine-tuning, image or audio pricing, or enterprise discounts.
Verification
The current dataset was last checked against all three sources on November 13, 2026. The date is shown on every page. When a provider announces a future price change, we store the effective date and the new price, and the tools switch over on that date.
What is an estimate
- Token counts: OpenAI counts in the token counter are exact for the selected encoding (we reproduce tiktoken and tested it against the library). Claude and Gemini counts, and the words-to-tokens ratios, are estimates based on character counts.
- Monthly cost assumes 30 days and a steady request volume.
- Batch discounts are applied only where we have verified them on the provider's page (currently Claude).
- Reasoning or thinking tokens, storage fees for caches, taxes, and regional pricing are not modelled.
Corrections
If a number is out of date or wrong, tell us at info@chelure.com and include a link to the provider page. We check corrections against the source and update the dataset.
Independence
We are not affiliated with Anthropic, OpenAI or Google. Product names belong to their owners and are used only to identify the services. We may show advertising on this site; ads do not influence the data.