TokenEggs

Provider note · checked 2026-11-20

OpenRouter free models: what :free gets you, and where it stops

Free models on OpenRouter exist, but they are not a separate free API. They are catalogue entries priced at zero that share the platform's capacity with everyone else — so the useful question is not “is it free” but “what does free cost me”.

A free model is a variant, not a service

There is no second API to sign up for. A free model is a variant of a model that also has a paid entry, and the variant is marked with the suffix :free at the end of the model ID. The docs present it as a catalogue variant: it shows up in the model list as its own entry, with its own pricing, context length and endpoints — which is why you can filter for it and call it exactly like any other model.

Because it is the same model behind a different entry, output quality does not change. What changes is the capacity you are allowed to use — and the fact that a provider can withdraw a free variant.

model: "<author>/<model>:free"

If you do not care which model answers, use the router

OpenRouter also publishes a free router that you call like a model of its own. It picks a free model for each request, filtered by the capabilities you require — image input, tool calling, structured outputs — and the model field in the response tells you which one actually answered.

The trade is control for convenience: availability varies with what is online, latency is worse at peak hours, and free models as a group carry lower rate limits than paid ones. For a prototype that is fine. For anything a user is waiting on, it is a coin flip you should be aware of.

Rate limits follow your credits, not your accounts

This is the part people get wrong. OpenRouter governs capacity globally, and its FAQ states that the free-model rate limits depend on how many credits you have purchased, and that creating additional accounts or API keys does not change them. The current thresholds are printed on the Limits page and in the FAQ, so read them there instead of trusting a number copied into a blog post last year — including this page.

QuestionWhat the official docs say
Is the request billed?No, for requests that a :free entry answers.
Do extra keys raise the cap?No. Limits are governed globally; more accounts do not help.
Is a free limit per model?The docs handle limits per model, so expect differences between them.
Can a free entry disappear?Yes. It is a variant, and variants get withdrawn.

Where the bill actually comes from

A free model sits on the same API key as the paid ones, and every request is billed per token for whichever model answered it. That is how a “free” setup ends up with a balance: the wrong model gets called, not a wrong plan.

Two ways it happens: calling a catalogue model without its :free suffix, and leaving a paid model in a fallback list so that a rate-limited free request is answered — and billed — by the next model in line. If zero is the requirement, keep the model list to free entries only, or call the free router by itself.

Five minutes before you build on it

  1. 01
    Filter the catalogue for free entries

    The models page can be filtered to zero-price entries; the list changes, so filter rather than bookmark model names.

  2. 02
    Copy the ID with its suffix

    The :free tail is part of the ID. Drop it and you are calling the paid entry at the paid price.

  3. 03
    Check the capabilities first

    Not every free model accepts images or tool calls. Ask for the capability you need rather than falling back to text after a failed request.

  4. 04
    Send one request and read the model field

    It tells you which model answered — the only reliable check that you stayed on a free entry.

  5. 05
    Handle 429 as a normal state

    On shared capacity a rate limit is expected traffic, not an outage. Retry with backoff, or fall back to another free entry.

Questions this page gets

Are OpenRouter free models really free?

Requests answered by a model entry ending in :free are not charged. The key itself is shared with paid models, so what decides the bill is which model answered — check the model field in the response rather than assuming.

Do extra accounts or keys raise the limit?

No. OpenRouter's FAQ says capacity is governed globally and the free-model rate limits follow the credits you have purchased. More accounts do not move the ceiling.

Why did a free model return a 429?

You hit the rate limit for that model, which on free entries is shared with every other user of it. A 429 is a cap, not a billing event and not a bug in your code.

Can I use free models in production?

Read the terms of the model's provider first, and assume availability is not guaranteed: a free variant can be withdrawn, and free capacity is the first thing to degrade when a model gets popular. Prototype on free, budget for paid.

How do I find the current free list?

Filter the models catalogue by price, or check the free-variant and free-router pages in the docs. Any list on a third-party page — this one included — is a snapshot, not the source of truth.

Read this in Chinese: OpenRouter 免费模型