Every LLM price,
on one honest scale.

Understanding LLM API pricing

Large language model providers bill by the token — a chunk of text roughly ¾ of a word in English (about 4 characters). Every request has two metered parts: the input (your prompt, system instructions, and any context you send) and the output (the text the model generates back). Providers publish separate per-token rates for each, and those rates differ by orders of magnitude across models — which is exactly why an apples-to-apples table is useful before you commit to one.

Why we normalize to “per 1M tokens”

Providers quote prices in different units — per 1,000 tokens, per million tokens, sometimes per character. TokenCost converts every model to a single unit, US dollars per one million tokens, so you can line up a budget model against a frontier model and compare the numbers directly. All figures on this page use that unit.

Input, output, and cached-input rates

• Input (prompt) price — what you pay per 1M tokens you send to the model. Long system prompts, retrieved documents, and chat history all count as input.
• Output (completion) price — what you pay per 1M tokens the model returns. This is almost always the more expensive of the two, often 3–8× the input rate, because generation is the compute-heavy step. If your app produces long answers, output cost usually dominates your bill.
• Cached input — many providers now offer prompt caching: if you resend an identical prefix (a fixed system prompt, a large document) within a short window, the repeated tokens are billed at a steep discount. The table shows each model’s own cached-input rate rather than a flat estimate, because the discount varies widely by provider.

Context window and tiers

The context window is the maximum number of tokens (input + output) a model can consider at once — larger windows let you send more documents or longer conversations, but sending more tokens costs more. We also tag each model with a rough tier — Budget, Standard, or Premium — as a quick signal of where it sits on the price/capability spectrum, so you can filter to the class you actually need.

How to use this page

Filter by provider, switch the sort (cheapest input, cheapest output, largest context, newest release), or narrow by release year. On a phone the table condenses to the essentials — tap any row to expand the full detail. Once you’ve found candidates, project a real monthly bill on the cost calculator, price an exact prompt with the token counter, find the lowest-cost options on cheapest models, put finalists head-to-head on the compare page, check capability against price on the benchmarks, or see every price cut and increase we've logged on price trends. For where the numbers come from and how often they refresh, see our methodology.

Frequently asked questions