Topic map

Last reviewed 2026-10-30

PromptSpend answers a handful of questions about what LLM APIs cost. Each section below names a question and the pages that answer it.

What does one model cost?

• [Every model, cheapest first](/models/): input, output and cached rates per million tokens, with the source and date on each.
• Each model page adds three monthly bills, the full rate card and its recorded price history.

Which of two models is cheaper for my work?

• [Head-to-head comparisons](/compare/): two models costed on the same three workloads. The most-requested pairs also carry a verdict per workload, the point where the answer flips, both price histories and a sourced benchmark.
• [The calculator](/): your own token counts, your own volume, up to four models side by side.

What does each company charge?

• [Providers](/providers/): every vendor's models in one table, how its prices are spread, which discounts it offers and its recorded price changes.
• [Free tiers](/free-tiers/): what each vendor gives away, quoted from its own terms and re-checked each morning.

How many tokens is this text?

• [Token counter](/token-counter/): counts in the browser, nothing uploaded.

What changed, and when?

• [Price changes](/changes/): every vendor repricing, with our own corrections labeled apart. Also as a [feed](/feed.xml) and as [JSON](/data/price-history.json).
• [Writing](/writing/): monthly price-movement reports and longer pieces.
• [Corrections log](/corrections/): mistakes we found in our own work.

Can I use the data myself?

• [Open data](/data/): the catalog and the full price history as JSON and CSV, MIT licensed.
• The public API, the MCP server for coding agents and the VS Code extension read the same catalog.

How are prices checked?

• [How prices are checked](/about/): where every price comes from and how it is re-checked.