FAQ — FreeModel

Cost and billing

Is FreeModel actually free?: Partly. Providers hand out free quota, and the router spends that before it spends anything of yours. When no free-quota model fits the task, the request continues on a paid model at its listed rate. fm-v1-lite is the only free tier: any account can call it and no subscription is needed. fm-v1-standard and fm-v1-pro need the matching monthly plan, and without it the call is rejected with a 402 rather than silently charged.
Where do I pay?: In the console, not on this site. This site is a catalogue and a reference; billing, subscriptions and balance live behind the account. Nothing here takes a payment.
What if I already pay providers directly?: Then bring your own keys. Add them in the console and that provider bills you directly; FreeModel adds nothing on top. Free models in the catalogue are still usable the same way.
Do free quotas expire?: Usually, and it is the provider's schedule, not ours. The most common shape is one million tokens valid for 90 days. We cannot extend a quota, restore one, or promise it will exist next month — so the router checks the live balance at request time.

Keys and access

Do I have to hand over my provider keys?: No. Use the platform key, or bring your own for the providers you already pay. Both routes go through the same router, so free-quota preference and failover behave the same either way.
How do I get a key?: Sign in to the console, open My → API Keys, create one. Keys start with sk-.
Can I look around before signing up?: Yes. /v1/models and the price catalogue are public and need no key. Worth doing if coverage is the thing you are checking.
Does it train on my traffic?: No. Requests are forwarded to the provider that serves them, and the routing log keeps what is needed to explain a call: which provider answered, which were skipped, why.

How routing behaves

What happens when a provider goes down?: The request moves to the next candidate automatically. A rate limit, an exhausted quota or a rejected key are all just triggers to try the next one. You get an answer instead of an error, and the log records what was skipped.
Will I know which model actually answered?: Yes, in the routing log. What you give up by using a route rather than a model id is deciding which one answers. If you need exact reproducibility, or a named model for compliance, send a concrete id instead.
Can a route change what it resolves to between calls?: Yes, and that is the point: it re-decides as providers add, retire and reprice models. Two identical requests a month apart can land on different models, which is why the log exists.

Clients and code

Which clients work?: Anything that speaks the OpenAI API or the Anthropic one. In practice: Claude Code, Codex CLI, Cursor, Cline, Continue.dev, and the OpenAI SDKs in Python and Node. Setup guides are under Docs.
Can I call it from my own code?: Yes, and it is one argument. Point the OpenAI SDK's base_url at https://freemodel.online/v1 and everything else stays the same: streaming, tool calls, usage reporting, error shapes. See the SDK guide.
Why do I get a 404 on every request?: Almost always the base URL. The OpenAI base ends at /v1 and the Anthropic base stops at /api/gateway; each client appends its own path, so pasting one into the other gives you /v1/v1/.... This is the single most common setup mistake; Docs has the exact strings.

FreeModel FAQ

How do I get an API key?

Sign in to the console, open My → API Keys and create one. Keys start with sk-. Copy the value when it is shown, because that is the only time the full string is displayed; if you lose it, create a replacement and revoke the old key rather than expecting the original to be recoverable.

Do I need an account just to look around?

No. /v1/models and the price catalogue are public and need no key. An account is needed to create a key and send a request, so checking coverage before signing up is the intended order rather than a workaround.

How is the free quota calculated?

It is not calculated by us — the allowance belongs to the provider. Each provider grants its own credits on its own schedule, and the router reads the live balance at request time rather than trusting a stored list. Free capacity is spent before anything of yours is, and when it runs out the same request falls through to the paid rate at the model's listed price.

What if I never want to be charged?

Send fm-v1-lite. It is the free tier and the only one that needs no subscription, so any account can call it. The fm-v1-standard and fm-v1-pro tiers do need the matching monthly plan; without it the call is rejected with a 402 rather than charged.

What happens when every candidate for a request fails?

You get a failure, not a silent substitution. Routing mitigates provider failure without removing it: a rate limit, an exhausted quota or a rejected key moves the request to the next candidate, and only once the candidates are exhausted does the call return an error. The routing log records which were tried and why each was skipped.

Can I use only my own provider keys?

Yes. Add them in the console and that provider bills you directly; FreeModel adds nothing on top. The request still goes through the same router with the same free-quota preference and the same failover, and the free-quota models in the catalogue stay usable either way.

Does FreeModel support the Anthropic API format?

Yes. The Anthropic-compatible base is https://freemodel.online/api/gateway, and it stops there because the client appends /v1/messages itself. Claude Code and anything else taking an ANTHROPIC_BASE_URL work without further changes. Pointing that base at /v1 instead is the usual cause of a 404 on every request.

How is this different from OpenRouter?

Both sit in front of many providers; the difference is what happens when free capacity runs out. OpenRouter publishes 460 public models and expects a funded balance behind them. FreeModel lists 200+ models. It spends provider free quota first, falling through to a paid model only when the route you sent allows it. If a specific model you depend on exists only there, verify before migrating — the comparison page covers both directions.

How is this different from LiteLLM?

LiteLLM is software you run: a library and proxy inside your own infrastructure, reaching providers with your keys. FreeModel is hosted. You point a base URL at it and nothing is deployed. Self-hosting keeps the network path and the key storage under your control; a hosted gateway gives you the routing order and the free-quota tracking without running either.

Do you store my prompts or the model's replies?

No. A request is forwarded to the provider that serves it, and what is recorded is metadata: model, token counts, HTTP status, duration, timestamp. The one exception is the agent endpoint, which stores the task text and the system prompt truncated to the first 500 characters. For sensitive material, use the standard chat and messages endpoints.

How do I delete my request data?

Call DELETE /agent/log with your key. That removes your rows from the request log, and it does not reach server-level access logs, which the hosting providers keep under their own policies. The privacy page has the call and its boundary in full.

Which clients work out of the box?

Anything that speaks the OpenAI API or the Anthropic one: Claude Code, Codex CLI, Cursor, Cline, Continue.dev, and the OpenAI SDKs for Python and Node. Setup guides are under Docs. Client support is rarely the limit[1]: 153 of the 199 listed entries are chat models, and chat is what an agent-style client calls.

What do the auto/ routes actually do?

A route describes the kind of output instead of naming a model: auto/image, auto/video, auto/tts, auto/asr, auto/embed and auto/rerank. The router filters the catalogue by modality, then prefers free quota, then falls back on failure. What you give up is knowing which model will answer; if you need a named model for compliance or reproducibility, send a concrete id.

How much of the catalogue can an agent actually use?

More than the raw model count suggests. Of the 199 listed entries, 153 are chat models; the other 46 are image, video, speech synthesis, speech recognition, embeddings and rerank, and they answer a different kind of request. An agent loop lives in the chat half, so verify against /v1/models, which is the authoritative list.

Do I need a payment method to start?

No. Free capacity needs no card, and the model list needs no key at all. Paid plans are bought on the account portal at alli.website, where the charge is collected by Creem; this site takes no payment itself and holds no balance.

Is there a rate limit?

Free capacity is rate-limited and shared, because it is a provider's allowance rather than a private one. It suits interactive use and is a poor fit for batch work. Paid models follow the provider's own limits, and a rate limit is one of the triggers that moves a request to the next candidate.

Connect a client in four steps

Step 1: Create a key in the console under My → API Keys. Keys start with sk-.
Step 2: Pick the base URL that matches the client: https://freemodel.online/v1 for anything speaking the OpenAI API, https://freemodel.online/api/gateway for anything speaking the Anthropic one.
Step 3: Set it through the environment rather than the app's own settings where you can: OPENAI_BASE_URL or ANTHROPIC_BASE_URL, with the key in OPENAI_API_KEY or ANTHROPIC_API_KEY. A base URL pasted into the wrong slot is the most common 404.
Step 4: Send one request, then open the routing log to see which model answered and which candidates were skipped on the way.

What the catalogue actually contains

Of the 199 entries the public endpoint returns, 153 are chat models, 18 are image generation and the remaining 28 are video, speech synthesis, speech recognition, embeddings and rerank. That is 77%, 9% and 23% of the list[1], and the split matters more than it sounds: the non-chat entries have no place in a text workflow at all, so the pool a chat client chooses from is the 153, not the 199.

Six of the entries (3% of the list) are modality routes rather than models — named policies such as auto/image and auto/rerank that resolve to a real model at request time, each carrying a candidates count that states how many models sit behind it.

How to use this FAQ without reading it all

Three questions cover most decisions: what a gateway is, whether the free tier is enough, and what happens when a provider fails. Start there; the rest are edge cases that matter only after you commit. Each answer states its scope, so a "yes" for a hobby project is not a "yes" for production.

Just evaluating: read the free-tier and setup answers.
Moving to production: read failover, billing, and data-path answers.
Comparing vendors: read the billing and key-ownership answers.

Sources

Every third-party figure quoted on this site was read from the vendor documentation linked below, on the date stated. None of these links are affiliate links and none of these vendors paid to be listed. If a number here disagrees with the page it came from, the page is right — tell us and it gets corrected.

Where the percentages come from

Every share quoted on this site is counted from the live model list, not estimated. Divide the two counts in the last column, round to the nearest whole percent, and you get the number in the first.

Share	Of what	Count
77%	are chat models	153 of 199 listed entries
9%	generate images	18 of 199 listed entries
23%	are not chat models	46 of 199 listed entries
3%	are modality routes	6 of 199 listed entries
17%	carry a context length	34 of 199 listed entries

[1] Counted on 29 September 2026 from GET https://freemodel.online/v1/models, which needs no key and no account. Every row uses all 199 listed entries as the denominator, because the only thing the endpoint states for every entry is its modality. It no longer publishes per-model capability flags such as tool calling or reasoning, so this site quotes no shares for them. Fetch the list and recount whenever you want to check the arithmetic.

Questions people ask before signing up