TOLVYN AI Cost Index

Methodology

The privacy floor, first, because it governs everything else. A figure is published only when at least three separate organisations contributed traffic to that exact bucket — same provider, same model, same period. Buckets below three are discarded during collection: they are never stored in a publishable form and never reach this page. The rule is applied before aggregation rather than as a filter afterwards, so there is no version of this index in which one customer's costs can be read off it. If your organisation were the only heavy user of a model, that model would simply not appear here — and that is why the index can legitimately show nothing.

• Source: Data is collected nightly from TOLVYN customers who have opted into data sharing.
• What we collect: Provider name, model ID, token counts, cost per request, request timestamp.
• What we never collect: Request content, prompts, completions, user data, company names, or any personally identifiable information.
• Minimum sample size: Each data point requires at least 3 organisations to contribute; anything below that is suppressed entirely rather than rounded, blurred, or noised. Suppression is the whole mechanism — we do not publish a degraded figure in place of a withheld one, because a degraded figure derived from one customer is still derived from one customer.
• How we calculate cost: Using the provider's published pricing at the time of each request, applied to the reported input and output token counts.
• Opt-out: TOLVYN customers can opt out of data sharing at any time from their account settings. Opted-out data is excluded immediately from the next nightly collection.
• Independence: TOLVYN is not affiliated with OpenAI, Anthropic, Google, or any AI provider. This index is published for informational purposes only.