OpenRouter Free Models Are Capped at 50 Requests a Day — Here Is How to Get 1,000 (2026 Setup Guide)

“OpenRouter has all these models ending in :free — can I just use them forever?”

No. And the wall is lower than most people assume: 50 requests per day. Run one slightly chatty agent and you will hit it before lunch.

The good news is that there is an officially documented way past it, and it costs a single one-time payment rather than a subscription. This guide is not about whether OpenRouter is worth using — that belongs on the pricing page. This is the operational side: signup to first request, the real condition that turns 50 into 1,000, and the four error codes you will definitely meet.

The short version

• Free models have three hard ceilings: 20 requests per minute, 50 per day (under US$10 purchased), 1,000 per day (US$10 or more purchased)
• That US$10 threshold counts credits purchased all time, not your current balance — this is the single most important line in this article
• If your balance goes negative, even free models are blocked (you get a 402)
• Every key can carry its own spend limit and reset cycle. Almost nobody sets this, and everybody should.

The three official numbers (read directly on 2026-08-26)

OpenRouter’s rate limit documentation states the free-model rules plainly:

The official wording for the threshold is “Credits purchased (all time)” of “At least 10” — note all time. Buy US$10 of credits, spend every last cent of it, and the 1,000-per-day entitlement stays.

That runs against the intuition most prepaid services train you into, which is exactly why it deserves its own line.

Our own earlier check of openrouter.ai/pricing (2026-07-31) recorded “25+ free models, 50 requests per day” — consistent with the official docs on the 50 figure. What our older record did not mention is the jump to 1,000 after US$10. That is the gap this article fills.

Step 1: Get a key (3 minutes)

1. Sign up at openrouter.ai (Google or GitHub sign-in both work)
2. Open the Keys page in your account settings and create a new key
3. The plaintext key is shown exactly once. The official docs say it is “returned only in this response” and “it cannot be retrieved later.” Copy it immediately.

When creating a key, the official API exposes these fields, which map to the web interface:

On reset timing the docs are specific: resets happen automatically at midnight UTC, and weeks run Monday through Sunday. If you are scheduling batch jobs, that boundary matters.

Step 2: Find which models are actually free

Free models are identified by the :free suffix on the model ID. You can filter on openrouter.ai/models, or pull the list programmatically from the official /api/v1/models endpoint and filter it yourself.

Pulling the list is the more reliable habit, because the free roster changes — a provider can withdraw the free version of a model at any time. Our 2026-07-31 record said “25+ free models across 4 free providers.” That was a snapshot, not a guarantee.

Step 3: Your first request

The endpoint is OpenAI-compatible, which is the genuinely useful part: code already written against the OpenAI SDK needs two lines changed.

The base URL from the official quickstart:

https://openrouter.ai/api/v1

Using the official OpenAI SDK, exactly as the docs show it:

from openai import OpenAI

client = OpenAI(
  base_url="https://openrouter.ai/api/v1",
  api_key="<OPENROUTER_API_KEY>",
)

completion = client.chat.completions.create(
  model="~openai/gpt-latest",
  messages=[{"role": "user", "content": "What is the meaning of life?"}]
)

Swap model for one of the :free IDs you found in Step 2 and you are running on the free allowance.

Raw HTTP works too: post to https://openrouter.ai/api/v1/chat/completions with an Authorization: Bearer <your key> header.

One note on that example: ~openai/gpt-latest is an alias, described officially as one that “always resolves to the newest OpenAI flagship model.” Convenient — but if output stability matters, pin an explicit model ID in production rather than an alias that moves under you.

Step 4: Turning 50 into 1,000

There is exactly one condition: US$10 in credits purchased, all time.

What that actually costs: OpenRouter charges a platform fee on credit purchases, recorded at 5.5% in our 2026-07-31 check of the official pricing page. So landing US$10 of usable credit costs slightly more than US$10 on the card.

The docs do not state whether the threshold is measured on the amount charged or the credits received after the fee. We could not find an answer, so we are flagging it as unverified. If you are topping up specifically to clear the bar, the safe move is to buy US$11 to US$12 rather than sitting exactly on US$10.

There is a subtler trap as well: a negative balance locks you out of free models too. The official wording is “If your account has a negative credit balance, you may see [402] errors, including for free models.” Note those last four words — including for free models. Even a free-models-only workflow cannot afford to leave a debt sitting on the account.

Step 5: The four error codes

From the official error reference, here are the four that actually show up when running free models (read 2026-08-26):

429 deserves extra attention. The docs state that rate limiting applies at both the request level and the token level — so “I did not even make 50 calls and I am being throttled” is entirely possible. Do not hardcode a retry delay; read Retry-After.

Three mistakes newcomers make

1. Confusing balance with lifetime purchases. Worth repeating: the US$10 bar is cumulative, and spending it down does not revoke the entitlement. Conversely, a US$50 balance made of promotional credit you did not buy may not count — the docs say “credits purchased.”

2. Wiring a key into an agent with no spend limit. One account reaches hundreds of models. limit plus limit_reset is free insurance; ten seconds at key creation beats an argument with your statement.

3. Assuming free models are permanent. The free tier is offered by providers, not promised by OpenRouter. Do not make a :free model the only path in production — keep a paid fallback.

Is the US$10 worth it just for the quota?

Split it out:

• Do not pay: 50 per day, 20 per minute. Fine for occasional testing, comparing model outputs, small scripts.
• Pay once (about US$10.55 including the 5.5% fee): 1,000 per day, permanently, and the credits themselves remain spendable on paid models.

So this is not a subscription. It is a one-time entry fee that also converts into usable credit. If you run agents or batch jobs on OpenRouter regularly, 50 calls runs out fast.

But if you only want to play with free models, do not rush to pay — Google AI Studio’s free tier is a no-card alternative with a different quota system entirely.

What we could not verify (stated plainly)

• How the US$10 threshold is measured. The docs say “credits purchased (all time)” but not whether that is the charged amount or the post-fee credit. Unverified.
• Whether free models require a data-sharing opt-in. The official privacy page says users can “set whether you would like to allow routing to providers that may train on your data” and that there are “separate settings for paid and free models” — but it does not name the specific toggle or where it lives. Not enough to write a step, so we did not write one.
• The exact number of free models today. It moves, and no count is promised. Query /api/v1/models yourself.

Sources

Rules and figures come from OpenRouter’s official documentation domain, read directly on 2026-08-26:

1. API Rate Limits — 20 RPM, 50/1,000 RPD, the US$10 all-time threshold, negative-balance 402
2. Error Codes — official wording for 402, 429, 404, 503 and the Retry-After header
3. Quickstart — base URL, OpenAI SDK compatibility, model aliases
4. Create API Key — limit, limit_reset, expires_at, midnight-UTC resets, one-time plaintext display

The free-model count and the 5.5% platform fee come from our existing checks of openrouter.ai/pricing (2026-07-31, 2026-08-22).

Limits and rules can change at any time; official announcements take precedence. Full plan and pricing detail is on our OpenRouter tool page; for another route to low-cost inference, see Gemini 3.7 Flash introductory pricing.