In November 2026, OpenRouter model IDs ending in :free cost $0 per token, with limits of 20 requests per minute and 50 requests per day. A qualifying credit purchase raises the daily allowance to 1,000. OpenRouter's free-model collection also lists other $0 models that can sit outside the :free caps.
OpenRouter free model limits
OpenRouter applies these limits to IDs ending in :free. Eligibility for the higher daily tier is based on all-time credits purchased, and OpenRouter says that extra accounts or API keys do not change the rate limits.
| Account | :free requests per minute | :free requests per day |
|---|---|---|
| Standard free tier | 20 | 50 |
| After a qualifying credit purchase | 20 | 1,000 |
Source: OpenRouter pricing, November 2026.
The higher tier applies once your account has bought at least $10 of credits, according to OpenRouter's pricing. Credit purchases on the Standard plan carry a 5.5% fee.
The daily counter covers the current UTC day, according to OpenRouter's limits documentation. A negative balance blocks free-model requests until you add credits, and failed attempts count toward the daily limit, according to OpenRouter's pricing page.
What does 50 requests per day buy you?
Say one agent task takes five :free model calls, including follow-up calls. At that rate, 50 divided by 5 gives you 10 tasks per day on the standard tier, while 1,000 divided by 5 gives you 200 tasks on the higher tier.
If a task takes 10 calls instead, the figures fall to 5 and 100 tasks, respectively. Retries can consume allowance too, so leave room for them rather than planning around every request succeeding.
The minute limit is separate. If each task needs five calls within the same minute, 20 divided by 5 gives you four tasks in that minute, even when plenty of daily allowance remains.
Which free models are on OpenRouter?
In November 2026, the most-used models on OpenRouter's free collection include Nemotron 3 Ultra (nvidia/nemotron-3-ultra-550b-a55b:free), Nemotron 3.5 Super VL (nvidia/nemotron-3.5-super-vl:free) and ERNIE-5.5-Flash-Preview (baidu/ernie-5.5-flash-preview:free). The collection ranks models by tokens processed over the trailing seven days.
The list changes. For a current selection, check the collection or pull OpenRouter's /api/v1/models catalog and inspect the :free IDs.
Free IDs can disappear from the catalog, so treat a listed expiry date as a cue to prepare a replacement.
A model can move from free to paid access too. Check the price before replacing a free ID with its paid counterpart. Our guide to free model IDs and expiry dates covers how promotional access can end.
The free router selects a model for you
OpenRouter's openrouter/free router selects an available free model at random. It filters for capabilities requested by the call, including tool calling, image understanding and structured outputs.
That saves you from pinning every request to one free ID. It also means two requests in the same conversation can go to different models from different companies, with different data policies. Some providers' published policies allow training on prompts, so review those policies before sending sensitive prompts.
Why free models return 429 errors
A 429 means a request was rate limited. On OpenRouter, that error can come from OpenRouter itself or from the upstream provider serving the model.
| Error | What it means | What to do |
|---|---|---|
Daily :free limit reached | The daily allowance for :free IDs has been reached. | Wait for the midnight UTC reset or use a paid model. |
Provider returned error with an upstream rate-limit message | The provider is rate limiting the model or is at capacity. | Try another model or retry later. |
| An OpenRouter rate-limit error | A :free model, key-management or other OpenRouter limit may have been reached. | Read the error and its rate-limit headers before retrying. |
Source: OpenRouter's limits documentation.
For an upstream error, error.metadata.provider_code can carry the provider's original error code when available. When OpenRouter itself rejects a free-model or key-management request, the error can include X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset headers. Successful responses do not carry those headers, according to OpenRouter.
Retrying every 429 the same way can waste time and allowance. OpenRouter says failed attempts count toward the daily cap. Check your own daily counter through GET /api/v1/key; the response includes free_model_daily_requests.
For more on provider errors and routing, see our guide to OpenRouter rate limits. Our 429 error guide covers rate limits across LLM APIs.
Set up a fallback before the free model fails
OpenRouter's models parameter lets you give a request an ordered list of models. If a model's providers are down, rate limited or refuse to reply, OpenRouter can try the next model.
The daily counter covers requests across :free IDs, according to OpenRouter's limits documentation. A free-only fallback list will not help after a daily-limit 429: wait for the reset or use a paid model.
- 1Pick a free model
Choose a current
:freeID from the live collection and check whether its capabilities fit your task. - 2Add alternatives
Use the
modelsparameter to list alternatives in the order you want OpenRouter to try them. A free-only list can help with provider capacity errors, but switching:freeIDs will not get around their daily cap. - 3Decide whether to pay for the last fallback
If the task must continue after the free options fail, add a paid model to the list. You will be billed at that model's paid price if OpenRouter uses it, so check the price before enabling it.
- 4Check the counter and errors
Read
free_model_daily_requestsfromGET /api/v1/keyand inspect 429 messages. A provider capacity error can clear sooner than a daily cap.
This free-only request body lists two models in order. OpenRouter can try Nemotron 3 Ultra if Nemotron 3.5 Super VL cannot serve the request:
{
"models": [
"nvidia/nemotron-3.5-super-vl:free",
"nvidia/nemotron-3-ultra-550b-a55b:free"
],
"messages": [
{
"role": "user",
"content": "Summarize the benefits of prompt caching."
}
]
}The next request adds the paid Nemotron 3.5 Super VL ID at the end. In November 2026, OpenRouter's model page lists Crusoe at $0.10 per million input tokens and $0.40 per million output tokens.
{
"models": [
"nvidia/nemotron-3.5-super-vl:free",
"nvidia/nemotron-3-ultra-550b-a55b:free",
"nvidia/nemotron-3.5-super-vl"
],
"messages": [
{
"role": "user",
"content": "Summarize the benefits of prompt caching."
}
]
}Take a paid call with 1,000 input tokens and 500 output tokens at Crusoe's listed rates. That comes to $0.0001 for input and $0.0002 for output, or $0.0003 per call before the credit purchase fee. The request body does not select Crusoe, so check the price of the provider that serves the paid call.
A paid model in the fallback list turns a free request into a billed request if OpenRouter routes to that model. Use only :free IDs when the job must stay free.
Free alternatives to OpenRouter
| Service | Published allowance |
|---|---|
| OpenRouter | In November 2026, 50 requests per day and 20 per minute for IDs ending in :free; 1,000 per day after a qualifying credit purchase |
| Requesty | In November 2026, 200 free-model requests per day and 20 per minute for new organizations |
| Cloudflare Workers AI | In November 2026, 10,000 Neurons per day |
| Cerebras | In November 2026, $5 in trial credits that expire after 30 days; a verified payment method is required |
Sources: OpenRouter's free-model collection, Requesty free-model docs, Cloudflare Workers AI pricing and Cerebras rate limits.
Cloudflare measures usage in Neurons, and the daily allowance is shared across models. Cerebras has no permanently free tier: its trial access stops when the credits expire or run out, unless you buy credits.
Requesty's free tier needs no credit card and has no trial expiry. Its limits apply across all free models combined and reset daily. Paying organizations get 1,000 free-model requests per day and 60 per minute, according to the Requesty docs.
If you upgrade to pay as you go, you keep the same API key and endpoint and change the model ID to a paid model. Requesty also supports fallback policies that try the next model when the primary fails on a timeout, rate limit or error.
Choose what happens when free runs out
For a side project that can wait, use free models and check the daily counter before retrying. For a task that needs an answer the same day, select a paid fallback and set a spending limit before you send traffic to it.
Start with Requesty's free models, then add a fallback policy when your application needs another route. Requesty has budget caps and per-user spend limits for paid usage.
Frequently asked questions
- How many free requests does OpenRouter allow per day?
- In November 2026, OpenRouter allows 50 requests per day for model IDs ending in :free, or 1,000 per day after a qualifying credit purchase. The higher tier applies once the account has bought at least $10 of credits.
- What is the OpenRouter free model rate limit per minute?
- In November 2026, OpenRouter model IDs ending in :free are limited to 20 requests per minute. They also have a daily limit of 50 or 1,000 requests, depending on the account's credit purchases.
- Why am I getting a 429 error on an OpenRouter free model?
- An OpenRouter 429 can come from OpenRouter's rate limits or from an upstream provider that is busy or rate limiting requests. Check the error message and provider metadata before deciding whether to wait or switch models.
- When does the OpenRouter free daily limit reset?
- OpenRouter's daily counter for model IDs ending in :free covers the current UTC day and resets at midnight UTC. The limit is 50 requests per day or 1,000 after a qualifying credit purchase.
- Does OpenRouter have a free model router?
- Yes. OpenRouter's openrouter/free router selects an available free model at random and filters for requested capabilities, such as tool calling or image understanding.
- What is an alternative to OpenRouter free models?
- Requesty gives new organizations 200 free-model requests per day and 20 per minute, with no credit card or trial expiry. Cloudflare Workers AI gives 10,000 Neurons per day, while Cerebras offers $5 in trial credits that expire after 30 days.
- JUL '26
OpenRouter rate limits: why 429s happen and how to fix them
OpenRouter rate limits in November 2026: free-model caps, why 429s happen, and how provider or model fallback and backoff can help you handle them.
- AUG '26
Free model IDs now ship with expiry dates: how to use promotional inference without building on sand
In one week: a stealth model gave away 20 trillion tokens then vanished, Thinking Machines put Inkling on OpenRouter free for a few weeks, MiniMax opened M3 and M2.7 free until 6 September, and Z.ai handed 50,000 developers 100M tokens each. Free inference is now a marketing channel with a calendar. Here is how to consume it without a production incident.
