OpenRouter Free Models: :free Variants & Rate Limits (2026)

Short answer: OpenRouter lets you call a rotating set of models for free. Any model ID that ends in :free costs nothing per token, and there is also a Free Models Router (model ID openrouter/free) that picks a free model for you. Free models are limited to 20 requests per minute and 50 requests per day. If you have ever bought at least 10 credits (about $10), the daily cap rises to 1,000 requests per day. No credit card is needed to start.

Rate limits checked against OpenRouter's official limits page and FAQ, and the free model count taken from OpenRouter's public models API, on 2026-10-17. The free lineup changes every few weeks.

What you get for free

OpenRouter is a single API that routes requests to hundreds of models from many providers. On 2026-10-17 its public models list had 470 models, and 16 of them were free variants, priced at $0 for both input and output. Free variants are separate model IDs with a :free suffix and their own rate limits. The OpenRouter FAQ also says all new users get a small free allowance to test the service, but it does not give an amount.

Context windows are the model's listed maximum; the provider serving the free variant may cap output length lower. Models come and go without notice, so build your app to handle a model disappearing (see the tips below).

The Free Models Router

If you do not care which model answers, set the model to openrouter/free. OpenRouter's Free Models Router guide says it picks a free model at random from the ones available and filters for models that support what your request needs, such as image input, tool calling or structured outputs. The response tells you which model actually answered. This is the easiest way to keep a demo running when individual free models are removed.

Rate limits in detail

These numbers come straight from OpenRouter's limits documentation and apply to every model ID ending in :free:

• The threshold counts purchases, not your balance. Buying 10 credits once and then spending them still keeps you on the 1,000-per-day tier.
• 9 credits is enough in practice. The docs say the higher daily ceiling starts one credit below the threshold, to absorb rounding and top-up fees.
• Days are UTC. The daily counter resets at midnight UTC.
• Paid variants have no platform cap. The docs say that if you hit a free-variant limit, you can switch to the paid version of the same model, which has no OpenRouter-level request cap.
• Providers can still say no. A 429 error can come from OpenRouter or from the upstream provider being at capacity. Free variants are often busy at peak times.

Checking how many free requests you have left

Send a GET request to the key endpoint. The response includes a free_model_daily_requests object with used, limit and remaining for the current UTC day, plus is_free_tier, which tells you whether the account has paid for credits before. The per-minute limit is not reported there.

curl https://openrouter.ai/api/v1/key \
  -H "Authorization: Bearer $OPENROUTER_API_KEY"

How to start, step by step

1. Create an account at openrouter.ai.
2. Go to Settings > Keys and create an API key. You can optionally set a credit limit on the key; for free models alone you do not need one.
3. Open Settings > Privacy and decide whether you allow providers that may train on your prompts. OpenRouter has separate settings for paid and free models, and some free models are only offered by providers that may train, so a strict setting can hide some of them.
4. Call a free model with the OpenAI-compatible chat completions endpoint (example below), or try it first in the Chat Playground by searching for free in the model picker.
5. If 50 requests per day is too few, buy 10 credits once on the Credits page to unlock 1,000 per day.

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "google/gemma-4-31b-it:free", "messages": [{"role": "user", "content": "Hello!"}]}'

Because the endpoint is OpenAI-compatible, most SDKs and tools work if you change the base URL to https://openrouter.ai/api/v1 and use your OpenRouter key.

Is the 10-credit purchase worth it?

For most developers who use free models every day, yes. The 10 credits are not used up by free models, so you keep them for paid calls later. What you pay on top is the purchase fee: OpenRouter's FAQ lists a 5.5% fee with a $0.80 minimum for card payments and 5% for crypto. On a $10 card purchase, 5.5% is $0.55, so the $0.80 minimum applies. In return, your daily free-model cap goes from 50 to 1,000 requests, a 20x increase.

• Refund window: unused credits can be refunded within 24 hours using the refund button on the Credits page. Platform fees are not refunded, and crypto payments are never refundable.
• Expiry: the FAQ says OpenRouter reserves the right to expire unused credits one year after purchase.
• Keep the balance positive. OpenRouter says a negative balance can cause 402 errors even on free models. This mostly affects people who also use paid models.

Privacy and data use on free models

OpenRouter's provider logging page explains that each provider has its own policy on logging and training. OpenRouter shows each endpoint's policy and lets you block providers that may train on your data, with separate switches for paid and free models. If you opt out, OpenRouter will not route to providers that train. That setting does not change what OpenRouter itself does with prompts, which is covered by its own data collection policy.

Free endpoints are often run by providers who are collecting feedback on new models. Do not send customer data, secrets or anything under NDA to a :free model unless you have checked that endpoint's policy.

Good uses for OpenRouter free models

• Comparing models before you pay. Run the same prompts through Gemma 4, Nemotron and Inkling to see which style suits your task, then switch to a paid model ID with the same code.
• Personal tools and scripts. 50 requests per day is enough for a daily summary script, a commit-message helper or a small chatbot for yourself. 1,000 per day covers most hobby apps.
• Coding agents on a budget. Coding-focused free models such as Poolside Laguna and Cohere North Mini Code can be plugged into tools that accept an OpenAI-compatible endpoint. Agents make many calls per task, so expect to hit 20 RPM quickly.
• Content moderation tests. NVIDIA Nemotron 3.5 Content Safety is free and can be used to prototype a safety filter.
• Teaching and workshops. Students can follow along with a free key and no card. For bigger classes, note that every account has its own 50-per-day cap.

How it compares with other free APIs

For a broader list, see our guide to free AI API keys. If you end up paying for frontier models through OpenRouter, compare against buying directly: OpenRouter says it passes through provider prices without markup, so the main extra cost is the credit purchase fee. See our OpenAI API pricing and Claude API pricing pages for direct rates.

Tips to avoid broken apps

• Never hard-code one free model. Use openrouter/free, or OpenRouter's model fallbacks feature to list a few models in order, so a removed model does not take your app down.
• Back off on 429 errors. Retry with exponential backoff and honor the Retry-After header when it is present. Hammering retries burns your 50 daily requests fast.
• Watch for errors mid-stream. If a rate limit hits after a streamed response has started, OpenRouter sends the error as a stream event with finish_reason set to error instead of an HTTP 429.
• Track usage with the key endpoint rather than counting requests yourself, because failed upstream attempts and retries can change the count.

Recent changes

• Free lineup churn: the set of :free models changes every few weeks. OpenRouter's own Free Models Router guide still mentions older examples such as DeepSeek R1 and Llama free variants, which were no longer in the models list on 2026-10-17.
• Free Models Router: openrouter/free was added to the catalog in early 2026 as a stable model ID for free inference.
• Limits: the 20 RPM, 50 per day and 1,000 per day after 10 credits rules were unchanged when we checked on 2026-10-17.