Runcrate is our editor’s choice for the cheapest LLM API: 125 models behind one OpenAI-compatible API, one prepaid balance, and the best developer platform we know.
The questions below are the ones readers bring to Cheapest LLM API, gathered from every entry on the site.
Before you call an LLM API the cheapest
Start here. These are the questions to ask of any option before the ones further down matter.
01
Does the model do the task?
A lower rate on a model that fails is not a saving. Test your own prompts on it first.
02
What is the output price?
Output costs more than input on nearly every listing. If your answers are long, the output price decides the bill.
03
What is your mix?
Count the input and output tokens in real requests. The mix tells you which of the two prices to weigh.
04
Can you use the free tier?
Read its limits: requests per day, tokens per minute and which models. A free tier that stops at noon is not a price.
05
What is behind the low rate?
Precision, context window and output limit differ between listings of one model. Read them beside the price.
06
When was it read?
Lowest listed means lowest on the day it was read. Check the date before you rely on it.
Questions and answers
Runcrate is our editor’s choice for the cheapest LLM API: 125 models behind one OpenAI-compatible API, one prepaid balance, and the best developer platform we know. Runcrate publishes this site.
It depends on the model and on your mix. For one model, the price table shows the lowest listed input and output prices on the date they were read, with Runcrate first.
Find the smallest model that does your task. Then measure your input to output mix and compare the price that carries your bill.
Other hosts’ prices are listed as text with the date they were read. This is a Runcrate reference, so its links go to Runcrate.
Only if it is the right model, the low price is on the token you use most, and its limits fit your volume. Check those three first.
Move the work that a smaller model can do to a smaller model. The gap between model sizes is wider than the gap between hosts.
Yes. Tokens a model spends thinking before it answers are output tokens, billed at the output price, whether or not you see them.
It lowers the bill, not the rate. You pay for the tokens written, so a shorter answer costs less.
There is none worth assuming. Retrieval and agents run many input tokens to one output token, and writing runs close to even. Measure yours.
No. You are billed the input and output prices separately. A blended price is an average at an assumed mix.
For testing, yes. For steady use, its limits decide. Once you pass them, the price is whatever you pay elsewhere plus the work of switching.
Because a free rate comes with a cap. It is a different product from a paid rate you can use at any volume.
No. It is the lowest published rate for one token type on one day. Your cost depends on your mix and on what the listing includes.
Without a schedule. Read the date beside any price, and read the price again before you commit to it.
It is a smaller one. On simple, high-volume tasks you may not see a difference. On hard tasks you will, so test before you move them.
The model. The price gap between sizes of one family is wider than the gap between hosts of one model.
No. The model still writes a fresh answer each time. Only the reading of the repeated opening is reused.
No. It lowers the price of repeated input only. Output is billed at the output price.
The window is a limit, not a charge. What costs more is filling it, because every token in it is billed as input on every request.
Nearly always. Sending the few passages that matter costs a fraction of sending the whole document each time.
One million divided by the price per million tokens. Work it out for input and for output, since the two prices differ.
By units of work per dollar. Measure the tokens one task uses on each model and divide the budget by the cost of that task.
The tokens you are billed, including retries and second passes, plus the time spent on limits and integrations. The token price is the first line.
Only if the saving at your volume is larger than the cost of the switch. One API with many models makes most switches a change of model name.
Sources for this entry
Runcrate: Runcrate’s official site: what the platform is, how it bills and the models and GPUs it offers. Read 12 Nov 2026.
Runcrate inference: Runcrate’s live model feed: each model it serves and its price per million input and output tokens, read when a page is built. Read 12 Nov 2026.
Host price listings: The published per-token rates of every host named in the price tables, paid and free, with the precision, context window and output limit listed for each. Read 12 Nov 2026.
Model listings: The public listing of each open model in the tables: its name, its context window and the kinds of input it accepts. Read 12 Nov 2026.
Checklists are original editorial guidance. Published prices do not establish availability.
This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.