Summary: OpenRouter gives you one API key and one endpoint for hundreds of AI models, and a few dozen of them can be used at no cost. The free tier is genuinely useful for learning and prototypes, but it comes with a hard daily request cap, a few billing traps, and privacy settings worth checking. This breakdown explains exactly what you get for free, what a one-time $10 credit purchase changes, and what the paid plans cost, based on OpenRouter's official pricing page and documentation checked on 2026-10-27.
What OpenRouter actually is (and why people use it for free models)
OpenRouter sits between your app and the companies that serve AI models. Instead of signing up with several providers and juggling several keys, you send standard chat requests to https://openrouter.ai/api/v1/chat/completions and choose the model by name. The API follows the OpenAI format, so the official OpenAI SDK works if you point its base URL at https://openrouter.ai/api/v1 and use an OpenRouter key.
The reason it shows up in so many hobby projects is the free catalogue. On the pricing page, the Free plan lists "25+ free models" from "4 free providers", with chat and API access included. The exact list changes often, and OpenRouter itself warns that free model availability "changes frequently", so treat any list of names, including the examples below, as a snapshot rather than a promise.
The free tier in numbers
Here is everything the official docs commit to for free models, in one place:
- 20 requests per minute on any model whose ID ends in
:free. - 50 requests per day if you have bought less than 10 credits in total.
- 1,000 requests per day once you have bought at least 10 credits (all time, not per month). Credits are denominated in US dollars, so this means a $10 purchase.
- Failed attempts count toward the daily cap, according to the pricing page FAQ. A loop that keeps retrying a broken request can burn through 50 requests quickly.
- A negative balance blocks free models until you add credits and bring the balance back above zero.
- Making extra accounts or keys does not raise the limit. The limits page says capacity is governed globally, though different models have different limits.
The practical takeaway: 50 requests a day is enough to learn the API, test a prompt or demo a side project to a friend. It is not enough for anything users depend on. The single most cost-effective move for a hobbyist is the one-time $10 top-up, which multiplies the free daily cap by 20 and leaves you with credit you can spend on paid models later.
How to use free models in three ways
1. In the browser, with no code
- Open the Chat Playground at openrouter.ai/chat.
- Click Add Model (or press Ctrl+K / Cmd+K) and type "free" in the search box.
- Pick Free Models Router to let OpenRouter choose a free model for each message, or pick any model with "(free)" in its name.
- Send a message. The reply shows which model answered. In OpenRouter's own walkthrough the router picked Solar Pro 3 (free); you may get something different.
One detail many people miss: the FAQ states that chatroom conversations are stored locally on your device and do not sync across devices. If you clear your browser data, they're gone unless you exported them from the chatroom settings menu.
2. One specific free model through the API
Create an API key in your account settings, then append :free to the ID of a model that has a free entry. The docs' example is meta-llama/llama-3.2-3b-instruct:free. Only models that list a free version support the suffix, and the free version is its own catalogue entry with its own context length and endpoints, so it may not behave identically to the paid version. The models page filtered to free is the place to check what exists today.
3. Let the Free Models Router pick
Set "model": "openrouter/free" in your request. The router looks at what your request needs (image input, tool calling, structured output), filters the free pool to models that support it, and picks one at random. The response's model field tells you which one answered. This is the most resilient option when a single free model is overloaded, but you give up control: two identical requests may be answered by different models, which makes it a poor fit for anything where you need consistent output.
The billing trap: fallbacks to paid models
OpenRouter lets you pass a models array so that if the first model fails, the next one is tried. That is handy, and the pricing FAQ confirms that failed attempts are not charged for model tokens. But the free-variant docs carry an explicit warning: if you put a :free model first and a paid model second, a rate-limited or unavailable free model hands the request to the paid one, and you are billed at the paid model's price.
To keep a request truly free, list only :free models in the array, or use openrouter/free on its own, which falls back only to other free models. If you're wiring OpenRouter into an automation tool, double-check this setting there too.
Free vs Standard vs Business: what each plan costs
OpenRouter doesn't sell monthly subscriptions for its self-serve plans. You buy credits, each request deducts its cost, and the plan decides the fee you pay on top of each credit purchase. Model prices are passed through from the providers without markup on every plan.
| Plan | Price / fee | Models and limits | Notable features |
|---|---|---|---|
| Free | $0, no platform fee | 25+ free models from 4 free providers; 20 requests/min and 50 requests/day | Chat and API access, activity logs and export, 5 workspaces, community support. No budgets, prompt caching or BYOK. |
| Standard (pay-as-you-go) | 5.5% fee on credit purchases; no minimum, no lock-in | 500+ models from 80+ providers; paid-model rate limits are passed through from providers | Budgets and spend controls, prompt caching, auto-routing, management API keys, email support. BYOK: $25,000 of list-price usage/month with no fee, then 5%. |
| Business (pay-as-you-go) | 8% fee on credit purchases; no contract or minimum | Same 500+ models and 80+ providers | Everything in Standard plus EU and US in-region routing, advanced notifications, workload identity federation, and up to 1,000 workspaces. |
Enterprise also exists, with contract pricing, invoicing, SSO/SAML and fee discounts, but it goes through sales. Prices exclude VAT/GST where applicable, and card, AliPay and crypto (USDC) payments are accepted on Standard and Business.
So when should you pay?
- Stay on free if you're learning, testing prompts, or running a personal tool that makes a few dozen calls a day.
- Buy $10 once if 50 requests a day keeps running out but you're still happy with free models. You'll pay the 5.5% fee on the purchase.
- Spend credits on paid models when you need a specific model, consistent output, or no platform-level request cap. OpenRouter says paid variants have no platform-level request cap; limits there come from the provider.
- Consider Business only if you need in-region EU or US processing or many separate workspaces. For most individuals and small teams, Standard is the logical paid step.
Credits: expiry, refunds and per-key caps
Three details from the FAQ are worth knowing before you buy anything. First, OpenRouter reserves the right to expire unused credits one year after purchase, so a large top-up you barely use is not a safe store of value. Second, refunds for unused credits can only be requested within 24 hours of the purchase, using the refund button on the Credits page; platform fees are not refunded and crypto payments are never refundable. Third, deleting your account loses any unused credits permanently.
If you share a key with a script, an agent or a teammate, set a credit limit on that key. You can check its remaining balance, usage and today's free-model counter (free_model_daily_requests) with a GET request to https://openrouter.ai/api/v1/key.
Editor's note
Who it's for: developers, students and tinkerers who want to try many models behind one OpenAI-compatible key, and small projects that can live with variable free capacity. Not for: anyone who needs a guaranteed model, steady latency or production reliability on a zero budget; OpenRouter itself describes free models as usually not suitable for production.
Gotchas I verified in the docs: the 50/day cap includes failed attempts; a models fallback array that includes a paid model can quietly bill you; credits can expire after a year and refunds close after 24 hours; and training-on-prompts is controlled by a separate privacy setting for free and paid models, so check both in your account's privacy settings if you'd rather not route to providers that may train on your data.
Common errors and fixes: 429 on a :free model usually means you hit 20/min or your daily cap, so back off, honour any Retry-After header, or buy the one-time 10 credits. 402 means insufficient credits; a negative balance triggers it even on free models, and on small balances it can also be a temporary in-flight budget hold that clears after the Retry-After wait. 401 is a missing, invalid or revoked key. 503 means no provider currently meets your routing requirements, and a 404 with "no endpoints for this model found" is what OpenRouter returns when a model has been deprecated, so pick a current model ID from the models page.
FAQ
Is OpenRouter really free?
The free models are, within the daily limits, and the Free plan has no platform fee. Paid models cost the provider's per-token price, plus a 5.5% fee when you buy credits on Standard. If you mix free and paid models in a fallback list, the paid part is billed.
Does buying $10 of credits mean I'm on a paid subscription?
No. There is no monthly subscription on the self-serve plans. The purchase simply raises your free-model cap to 1,000 requests a day and leaves you with credits to spend on paid models whenever you want.
Can I plug OpenRouter into a chat app or automation tool I already use?
Usually yes, because it speaks the OpenAI format. Tools that accept a custom OpenAI-compatible base URL, such as the self-hosted setup in our Open WebUI team playbook, can point at https://openrouter.ai/api/v1. If you hit rate-limit errors inside a workflow, our n8n AI Agent errors guide explains how 429s show up there, and if you'd rather avoid caps entirely by running models locally, see our Ollama troubleshooting guide.
Start with the Free Models Router in the browser, move to a :free model ID in code, and only spend money once you know which model you actually need.
See the official screenshots at the end of this post. Image credit: OpenRouter Docs (openrouter.ai/docs) and our own screenshot of the public OpenRouter pricing page.
Last verified: 2026-10-27
Sources:
OpenRouter Pricing
OpenRouter Docs: Limits
OpenRouter Docs: FAQ
OpenRouter Docs: Free Variant
OpenRouter Docs: Free Models Router
OpenRouter Docs: Free Models Router in the Chat Playground
OpenRouter Docs: Quickstart
OpenRouter Docs: Errors and Debugging
OpenRouter Docs: Provider Logging
댓글
댓글 쓰기