LLM API pricing

What each model actually costs per million tokens.

Rates read off each provider’s own pricing documentation, stamped with the date they were checked, and cross-checked on every deploy against the live model catalog Spendline uses to price real API traffic. If the two ever disagree, this site fails to build rather than showing you a stale number.

Looking for what Spendline itself costs? See our plans.

By model

Head to head

When these prices last moved

Providers change rates and the old number disappears. We keep a dated log of every price change we have observed, including the ones that have been announced but have not taken effect yet.

Why these numbers are not the whole bill

Every page here shows a rate card, and a rate card is the smallest part of what a team actually pays. Output usually costs several times input, so responses drive the invoice even when prompts look larger. Reasoning tokens bill as output, which means raising an effort setting raises the bill without changing a prompt. Retries and agent fan-out pay full price for calls that produced nothing. And a provider invoice arrives as one number for the entire organization, so none of it is attributable to a customer or a feature by default.

The guides go into that in depth: what Claude actually costs and why the token estimate comes in low, how to track LLM costs per customer, and LLM unit economics.

Rates on this page last verified August 12, 2026.