Terms — the deal when you use FreeModel

What we provide

FreeModel forwards your API requests to third-party model providers and returns their responses. We do not host models, do not train models, and do not modify responses. The models are the providers' — we route to them and hand back what they return.

The public model list at /v1/models is the authoritative answer to what is callable right now. If a model is not on it, it is not available, regardless of what any other page says.

What free capacity actually is

Free capacity belongs to the providers, not to us. Providers grant allowances to win developers: token credits on signup, promotional quota, rate-limited plans. The router tracks those allowances and spends them before touching anything you pay for.

Three consequences follow, and they are the honest version of "free":

We cannot extend an allowance, restore one, or promise it will exist next month.
A provider can change or withdraw its allowance without notice to us or to you.
Free capacity is rate-limited and shared. It is fine for interactive use and a poor fit for batch workloads.

If your work depends on a specific model being available at a specific moment, do not build on free capacity alone.

No uptime guarantee

There is no SLA. The gateway is offered as-is, and routing mitigates provider failure but does not eliminate it. When every candidate for a request is unavailable, the request fails, with an error that says so rather than a silent substitution of a different model.

Your account and keys

Keep your API key secret; anything done with it is treated as done by you. If you bring provider keys of your own, you remain responsible for the terms of those providers and for the usage they bill you.

Do not use the service to break the law, to attack the providers behind it, to resell capacity in a way that abuses a free allowance, or to circumvent rate limits by rotating accounts. Access can be suspended for any of those.

Your content

What you send stays yours. We claim no licence over your prompts or the responses, and we do not store either; see Privacy for the one exception. The provider that serves your request processes your content under its own terms.

Liability

We are not liable for the output of a model, for a provider's decision to refuse or rate-limit a request, or for losses arising from a request failing. Use the service for anything load-bearing at your own risk, and keep a fallback path.

Changes, and how to reach us

These terms may change as the product does; the date below moves when they do. Continuing to use the service after a change means the new version applies. If a change is material, it is noted on the about page.

Questions: postmaster@gladec.cn. Grievances first, please. Most things are a bug rather than a dispute.

Where free capacity stops

"Free first" describes the router's order of preference, not a property of your account. For any given request the router filters the catalogue by what the task needs. A request that passes tools rules out models without tool support, and an attached image rules out models without vision. It then spends whatever free allowance the survivors have. Whether a free candidate exists at all therefore depends on the shape of the request as much as on the remaining balance. The list holds 153 chat models out of 199 entries, so a chat request rarely has a single candidate; a request that needs an image, a video or speech draws from a much smaller pool, which narrows the field far more sharply.

Three limits follow, and none of them is ours to relax. Free capacity cannot be reserved, prioritised or held for you: it belongs to the provider and is shared with everyone else drawing on that allowance. It cannot be refilled. A request that arrives after the balance reaches zero is simply a request with no free candidate. And it cannot be relied on for throughput, because the same allowance is rate-limited for everyone using it.

Spend control does not depend on free capacity. The tier aliases route by pool: fm-v1-lite is callable by any account and is not checked against a subscription. fm-v1-standard and fm-v1-pro require their monthly plan, and a call sent without it is refused with a 402 rather than billed. A concrete model id is billed at its listed rate.

Coverage is rarely the constraint. The public list is 200+ models: 153 chat models make up 77% of it, and the rest spread across image, video, speech, embeddings and rerank. Verify against /v1/models before depending on any single entry, because that list is the authoritative answer and it changes.

Terms FAQ

Is "free" a promise?

No, it describes the order the router spends in. Providers grant allowances to win developers and can withdraw them without notice to us or to you. What is promised is the ordering (free capacity is tried first), not the availability of any particular allowance.

What happens if nothing has free quota left?

On the standard routes the request continues on a paid model at its listed rate. The fm-v1-lite tier is callable by any account without a subscription. fm-v1-standard and fm-v1-pro need their monthly plan and are refused with a 402 without it. Which tier is set is the main spend-control lever on the API side.

Do you guarantee a model stays available?

No. Providers add, retire and reprice models on their own schedule, and we are told no earlier than anyone else. /v1/models is the authoritative answer to what is callable right now; if a model is not on that list it is not available, whatever another page says.

Can I build a commercial product on this?

Nothing in these terms restricts commercial use. What is restricted is using the service to break the law, to attack the providers behind it, to resell a free allowance abusively, or to get around rate limits by rotating accounts — any of those can end in suspension.

What are my obligations if I bring my own provider keys?

You stay responsible for those providers' terms and for what they bill you. Your key remains yours and the provider bills you directly; our part is the routing in between. Keep your API key secret in every case. Anything done with it is treated as done by you.

If a model returns something wrong or harmful, whose problem is it?

The output is the provider's model output and we do not modify it. These terms exclude liability for model output, for a provider refusing or rate-limiting a request, and for losses arising from a request that fails. Keep a fallback path for anything load-bearing.