In short
OpenRouter's free models, those with IDs ending in :free, are limited to 20 requests a minute. You get 50 requests a day until you've bought at least 10 credits, and 1,000 a day after that. They're good for learning, prototypes and testing several models. They're poor for production, since availability varies and providers may treat data differently. Don't send customer data without checking the terms.
What's actually on offer
OpenRouter is a single API in front of many AI models from many providers. Some of those models are free to use, with IDs ending in :free. It uses an OpenAI-compatible format, so existing code usually needs only a new base URL, API key and model name.
The limits
From OpenRouter's documentation, checked on 4 November 2026:
| Credits purchased, all time | Requests per minute | Requests per day |
|---|---|---|
| Fewer than 10 | 20 | 50 |
| At least 10 | 20 | 1,000 |
So one small one-time purchase raises your daily limit by 20 times. Community write-ups note that limits are per account, so extra keys or accounts won't help, and creating accounts to get around limits may breach a service's terms anyway.
You might also get a rate-limit error from the upstream provider, even if you're under your own limit, since free capacity is shared, and popular models get busy.
Which free models are available?
The list changes often. Models come and go, and some appear for a limited promotion. Check OpenRouter's model page, filtered by free, rather than relying on any article or a hard-coded list. It also means you shouldn't build anything that depends on one particular free model being there next month.
What they're good for
Learning. You can build your first AI feature without any cost, including streaming and basic tool use.
Prototyping. Test an idea end to end, and only pay once it's worth pursuing.
Comparing models. Try one prompt across several models, with one API, before choosing. Our guide to how to choose an LLM for your product explains what to look for.
Personal tools. A small script for your own use, at well under 1,000 requests a day, can run happily for free. Cost it up: if a script makes 200 requests a day, then it's well within the limit once you've bought credits.
Coding agents, within reason. You can connect them to some agent tools, but agents send many requests per task, so 50 a day disappears quickly, and even 1,000 can go fast. Model quality and tool-calling support also vary. See OpenCode vs Claude Code.
The catches
Reliability. Free capacity gets busy, and slowdowns or errors are common at peak times. There's no service guarantee, which is a poor basis for anything customers depend on.
Data handling. This is the one to take seriously. A free model may be free because the provider wants data for training or improvement. OpenRouter has settings to filter providers by their data policies, and each model page lists details. Assume that anything you send to a free endpoint may be logged, unless you've checked otherwise. Don't send customer data, private code or anything confidential without reading the terms. See is it safe to send customer data to an LLM?
Changing models. A model that disappears breaks your app. If it changes version, your prompts can behave differently. Always have a fallback.
Feature differences. Tool calling, structured output and context length vary from one model to another, and so does the quality.
Smaller limits for vision and long documents. Some free endpoints may have shorter context or smaller outputs than the paid version of the same model. Check before relying on them.
Handling the limits in code
If you do use free models, build for failure from the start:
import time
from openai import OpenAI
client = OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key="YOUR_KEY",
)
MODELS = ["first-free-model:free", "second-free-model:free"]
def ask(prompt):
for model in MODELS:
for attempt in range(3):
try:
r = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": prompt}],
)
return r.choices[0].message.content
except Exception:
time.sleep(2 ** attempt)
raise RuntimeError("All models failed")
Replace the placeholders with models currently available. In production, you'd catch specific errors and treat rate limits differently from other failures. See how to handle LLM failures gracefully.
When to start paying
Move to paid models when any of these apply:
- Real users depend on it. Paying customers shouldn't be subject to free-tier limits.
- You're handling private data. Choose a provider and terms that suit it.
- You've found a model that works, and you need it to be stable.
- You hit limits regularly. Paid models are often cheaper than people expect.
To put that in perspective, DeepSeek's Flash model lists at $0.15 per million input tokens off-peak. A workload of 10 million input and 2 million output tokens costs about $2.70 a month. Our LLM API pricing comparison shows several options around that level, and the cheapest way to run an LLM covers local alternatives, which suit private data on a low budget.
OpenRouter can also make sense as a paid gateway. Using one API for many models can simplify your fallback setup. Compare any fee it adds with going to a provider directly, and check whether caching and data terms suit you. Our SolCard Inference vs OpenRouter comparison looks at one alternative.
The verdict
Yes, they're worth using, for the right jobs. They're an excellent way to learn, prototype and compare. Buy the $10 of credits if you plan to use them regularly, because that one purchase gives you 20 times as many daily requests. Just don't put them in front of customers or feed them anything you wouldn't want stored.
If you're building something real on top of AI models, book a call.
Common questions
How many free requests does OpenRouter allow?
Its documentation lists 20 requests per minute on models ending in :free, with 50 requests per day if you've bought fewer than 10 credits and 1,000 a day if you've bought at least 10. Check the current limits, because they may change.
Are OpenRouter's free models safe for private data?
Don't assume so. Free endpoints may log prompts or use them to improve models, depending on the provider. Check each model's data policy and OpenRouter's settings, and avoid sending customer data or confidential code to free models.
Do I need a credit card for OpenRouter's free models?
You can use free models with the lower daily limit without buying credits. To get the 1,000-a-day limit, you'll need to have purchased at least 10 credits in total. Check current sign-up requirements on OpenRouter's site.
Can I use OpenRouter's free models in production?
It isn't recommended. Daily limits, shared capacity and models that can disappear make them unreliable for real customers. Use them for prototypes, and move to paid models when anyone depends on the result. Paid budget models are often very cheap.