What the OpenRouter Free Tier Actually Gives You
OpenRouter's free tier allows 20 requests per minute and 50 requests per day with no payment at all, and that daily ceiling jumps to 1,000 requests per day once an account has bought $10 in credits at any point in its life.
That second number is the part most coverage gets wrong. The 1,000 per day tier is not a subscription. It follows lifetime credit purchases rather than current balance, so a single $10 top-up keeps the higher ceiling after the balance is spent to zero.
| What you pay | Requests per minute | Requests per day |
|---|---|---|
| Nothing | 20 | 50 |
| $10 in lifetime credits | 20 | 1,000 |
These limits apply to models carrying the :free suffix. Around 19 models carried that permanent suffix at the time of the sources behind this page, including NVIDIA Nemotron 3 and members of Google's Gemma family. That count is not fixed: providers rotate models in and out, so the live catalogue is the only reliable answer. AI Perks tracks free tiers and credit offers across the model vendors those routes point at.

What Is Confirmed About Paid Usage, and What Is Not
Past the free tier, the one confirmed billing mechanic is that a request is charged through the cost of the underlying models it calls. Everything else in OpenRouter's commercial pricing is either unpublished or unverified, so no figure for it appears here.
OpenRouter coverage is full of confidently stated numbers that do not trace back to the vendor: top-up fee percentages, bring-your-own-key allowances, Business plan prices. Repeating them would make this page look more complete and your cost model worse.
| Claim you will see elsewhere | Status on this page |
|---|---|
20 req/min and 50 req/day on :free models | Confirmed |
| 1,000 req/day after $10 in lifetime credits | Confirmed |
| Requests billed through the underlying models called | Confirmed |
| A percentage fee on credit top-ups | Not confirmed, not quoted |
| A bring-your-own-key monthly allowance | Not confirmed, not quoted |
| A monthly price for the Business plan | Not published |
| A flat price for Fusion | Not published |
The practical consequence is a habit rather than a number: read OpenRouter's own pricing and FAQ pages before committing to a cost model, and treat any article quoting an exact fee percentage as unverified until the vendor's own site shows it.
What Stripe Buying OpenRouter Changes for Free Users
Stripe agreed to acquire OpenRouter in a deal reported in the $7 billion to $7.5 billion range, announced publicly on 19 August 2026, and nothing about the free tier changed at announcement.
Both companies published their own accounts of the deal. OpenRouter continues operating its marketplace of 400+ models under its own brand rather than folding into Stripe's product line.
Two things did not happen alongside the deal, and both matter if you are budgeting:
- No new startup credit program was announced with the acquisition.
- No change to free tier limits was published at announcement.
Acquisitions of this size do eventually reshape pricing, and a payments company owning a metered inference gateway has obvious reasons to touch billing. But that is speculation, not published policy. The practical hedge is to not build a product whose margin depends on one vendor's free tier surviving. AI Perks tracks the current terms alongside $7.7M in credits across 194 companies, which is the sort of buffer that survives a pricing change.

OpenRouter Fusion: A Panel of Models Behind One Prompt
Fusion is a compound model: a single prompt is dispatched to a panel of several underlying models in parallel, a judge model maps where they agree and disagree, and a final model synthesises one answer.
OpenRouter's blog index shows Fusion launching on 12 June 2026, with a dedicated explainer following on 10 September 2026. Those post dates are what the index confirms; the individual posts were not re-fetched for this piece, so read the description as reported rather than independently verified. The pitch is that a panel plus a judge beats single-model frontier performance on the same prompt.
Billing follows the panel. A Fusion request is charged through the cost of the underlying models it calls, and OpenRouter has not published a separate flat Fusion price. A Fusion request therefore costs a multiple of a single-model request, roughly in proportion to how many models sit on the panel.
Whether Fusion can draw on :free models, and therefore whether it is usable on the free tier at all, is not confirmed in the sources behind this page. Treat it as a paid feature until the vendor says otherwise. If you are experimenting, vendor credits for the models on the panel are the cheaper way to fund it.
US and EU In-Region Routing Answers the Compliance Objection
OpenRouter added a US and EU in-region routing option on 9 September 2026, described as guaranteeing that a request is decrypted and served only by providers physically located inside the chosen region.
This is a data residency feature rather than a latency one, and the distinction it draws is the one a compliance review actually asks about: not just where data was stored, but where it was decrypted and where inference ran.
The covered model list is reported to span vendors including OpenAI, Anthropic, Google, NVIDIA, Thinking Machines, DeepSeek, Moonshot and Qwen. No separate surcharge for the feature appears in the sources behind this page, and whether it is available on free tier requests is not confirmed. This comes from OpenRouter's blog index rather than a re-fetched product page, so confirm the terms with the vendor before citing them in a compliance document.
Even hedged, it is the shape of an answer to the compliance objection that has kept EU teams off routing layers.

How OpenRouter's Free Tier Compares to Other Free Inference
OpenRouter's free tier is request-capped, while several rival free tiers are token-capped, and that difference decides which one suits your workload.
Fifty requests a day is generous for an agent that makes a handful of long calls and restrictive for anything that fires many short ones. A token-based allowance inverts that.
| Provider | Free allowance | How well sourced |
|---|---|---|
| OpenRouter | 20 req/min, 50 req/day, 1,000/day after $10 lifetime credits | Confirmed in the sources behind this page |
| Cerebras | Reported at 1,000,000 tokens/day, resetting daily, no card required | Reported by third-party trackers only, not confirmed on the vendor site |
| xAI / Grok | Not quoted here | Public figures disagree across sources, so none appear on this page |
The Cerebras figure is worth flagging precisely because it is widely repeated and thinly sourced. Its GPT-OSS-120B pricing also appears inconsistently across public sources, and the vendor's own pricing table did not resolve the conflict on review, so no rate for it is quoted here either. Treat both as reported rather than confirmed and check the vendor directly.
For xAI, no per-token figure appears here on purpose. Published third-party numbers disagree enough that repeating any of them would be misleading. Their own console is the source of truth.
How to Fund OpenRouter Usage With Free AI Credits
A request is charged through the models it calls, so credits from the underlying model vendors are what actually reduce an OpenRouter bill, not credits from OpenRouter itself.
That is the structural point almost every "free OpenRouter credits" article misses. The gateway sits in front of the labs, so cutting model spend means holding credit from the labs you route to.
Step 1: Start at getaiperks.com and filter to the AI model vendors you are actually routing traffic to. Credit from a vendor you never call is worth nothing.
Step 2: Buy $10 in OpenRouter credits early. It takes a free account from 50 to 1,000 requests a day, and because the unlock follows lifetime purchases, it holds after the credit is spent.
Step 3: Separate prototyping from production. Prototype on :free models inside the request cap, then move production traffic onto credit-funded paid models.
Step 4: Re-check quarterly. Post-acquisition pricing is the single most likely thing on this page to change, and AI Perks tracks the terms as they move.

Frequently Asked Questions
Is the OpenRouter free tier actually free?
Yes. Models carrying the :free suffix are usable at no cost, limited to 20 requests per minute and 50 requests per day. Paid models and Fusion are billed through the underlying models a request calls. Free credits from model vendors are tracked at getaiperks.com.
How do I get 1,000 requests per day on OpenRouter?
Buy $10 in OpenRouter credits at any point. The higher daily ceiling is tied to lifetime credit purchases, not to a current balance or a subscription, so it persists after the credit is spent.
Does OpenRouter charge a fee on top of model prices?
Specific fee percentages circulate widely, but none of them are confirmed by the sources behind this page, so no figure is quoted here. What is confirmed is that a request is charged through the cost of the underlying models it calls. Check OpenRouter's own pricing and FAQ pages before modelling costs, and treat any article quoting an exact percentage as unverified.
Did Stripe's acquisition change OpenRouter pricing?
Not at announcement. Stripe agreed to acquire OpenRouter in a deal reported in the $7 billion to $7.5 billion range, announced 19 August 2026, and OpenRouter continues running its 400+ model marketplace under its own brand. No free tier change and no new credit program were published alongside the deal.
Can I use OpenRouter Fusion on the free tier?
Not confirmed. Fusion is billed through the cost of every underlying model it calls, and OpenRouter has not published whether :free models can sit on a Fusion panel. Assume it requires paid models until the vendor states otherwise. Fund experiments with free vendor credits.
Is EU data residency available on OpenRouter?
An EU and US in-region routing option was announced on 9 September 2026, described as keeping decryption and inference with providers physically inside the chosen region. Whether it reaches free tier requests is not confirmed, and no separate surcharge appears in the sources behind this page. Confirm the current terms with OpenRouter before relying on it.
Route to any model. Pay for as few of them as possible.