FreeModel is an AI model gateway that bills per token: you pay for the input and output tokens of the model that answered, at the rate in the tables below. A free tier sits in front of the paid catalogue — free-eligible models are tried first on every request, and paid capacity is the fallback.
Free tier, in numbers
Card required: No. Signup takes an email; no credit card is asked for before your first call.
Routing policy: Free capacity is tried first on every request; paid capacity sits behind it as automatic fallback. You do not tag individual calls.
Model coverage: 200+ models in the public list, behind one base URL. Free-eligible models include GLM-4.7-Flash, GLM-4.5-Flash, Hunyuan-lite, Nex-N2-Pro and the Yi series — all listed as Free in the table below.
Quota: Free capacity is drawn first. Your account's current allowance and reset window are shown in the console, because they differ per account and per tier.
At the limit: Requests fall through to paid capacity instead of failing outright. Paid capacity depends on your balance or subscription.
What counts as a billable call
A billable call is one request that reaches a paid model. Requests served from free capacity — the free-eligible models listed as Free in the table below — are not billed. One request is one call, regardless of how long the prompt is; the table prices are per 1M tokens, so a call's cost is its input and output token counts priced at those rates.
Routing is automatic. You send one request to one base URL. Free capacity is tried first; paid capacity is the fallback. There is no per-call switch to set.
You are billed for tokens, not for time. A long conversation costs what its tokens cost.
Failed calls from a provider are retried. If a provider errors out, the gateway falls over to the next candidate instead of returning the error, so a failure on the vendor side is not billed as a completed call.
How to check what you have left
Open the console and sign in with the account you use for the gateway.
Read the usage panel: it shows the free allowance already drawn and the reset window for your account tier. The number is per account, not a site-wide constant.
Compare it with the model list: every row marked Free is served from free capacity first, so a request to those models draws the allowance before anything paid is touched.
If you need a hard ceiling, set a budget in the console. When it is reached, paid capacity stops instead of continuing to draw from your balance.
How one request is billed, step by step
The request arrives at the OpenAI- or Anthropic-compatible base URL with your gateway key.
The gateway resolves the model name to a candidate list and puts free-eligible candidates first.
The request is sent to the first candidate. If that provider returns a quota error, a rate limit, or a hard failure, the next candidate is tried.
If a free candidate answers, nothing is billed. If a paid candidate answers, the input and output tokens are priced at that model's row in the table.
Usage is written to the console, where it is visible per key and per day.
Where these numbers come from
Model prices are taken from the providers' published rate cards and re-checked when a provider changes them. The primary sources for the rows above:
Zhipu — GLM rates, including the Flash tiers listed as Free.
Which tier fits which task
Light drafting, file cleanup, and occasional Q&A fit the free tier: they are bursty and easy to retry. Daily coding-agent work, long documents, and anything with a deadline fit Standard: the difference is not access, it is queueing and model tier. Pro is for parallel long-running tasks where a failed run costs more than the subscription.
No. Signup takes an email. A card only matters once you are drawing on paid capacity.
What happens when the free allowance runs out?
Requests fall through to paid capacity rather than failing. If you have no balance and no subscription, the request returns an error instead of quietly charging you.
Are the prices in USD?
Yes. Every figure on this page is USD per 1M tokens, matching what the upstream provider charges for that model.
Does routing to a cheaper model change what I am billed?
You are billed for the model that actually answered, at that model's rate. If a request fails over from an expensive candidate to a cheaper one, the cheaper row applies.
How current is this table?
Prices are pulled from provider catalogues and re-checked when a provider changes them; the date at the top of the page is the last time this table was verified. Model availability changes more often than prices do — the live list is always at /v1/models.
Free capacity vs paid capacity
Two pools answer the same request. Free capacity comes from models that upstream providers offer at no charge; it is drawn first and it is not billed. Paid capacity is everything else, priced by the tables below. You do not choose between them per call — the gateway decides, and only the model that actually answers is billed.
How to read the table
Columns are input, output and cached-input price per 1M tokens, in USD. A dash means the provider publishes no separate rate for that column. "Free" in a row means the model is served from free capacity and costs nothing while that capacity lasts.