What OpenRouter's Free Tier Gives You in November 2026
OpenRouter's free tier is still alive in November 2026, but it is not the same list you may have bookmarked in the spring. A daily check on 3 November 2026 listed 14 free text models and found only 7 of them answering that day, while a brand-new free stealth model joined the roster the same day. If you build prototypes on free endpoints, the lesson is simple: pick from what is answering now, keep a fallback, and never hard-code a single free model.
First, the important distinction
OpenRouter is a free API route, not a no-sign-up chatbot. You need an OpenRouter account and an API key, and you call it from code or an OpenAI-compatible client. That is exactly why it belongs on our free APIs page rather than alongside the no-sign-up chatbots in our agents directory. If you only want to chat in a browser with no account at all, this article is not your shortcut — it is for people who want free model access behind an API.
What the free tier actually allows
The published free-tier limits are 20 requests per minute and 50 requests per day on models with the :free suffix. The daily limit rises to 1,000 requests per day once the account has bought $10 of credits at any point — a one-time lifetime threshold, not a subscription, according to the tracker that re-checks these pages daily. No credit card is needed to start, but there are two traps worth knowing before you rely on the headline numbers.
A negative balance blocks even free models. If your credit balance drops below zero, OpenRouter returns a payment-required error for :free requests too, until you top the balance back above zero.
Your quota is rarely the failure. Each free model is served by an upstream provider with its own capacity. When that provider is saturated, you get an upstream rate-limit error even if you have made only one request all day. That was the most common failure in the tracker's tests, and it is the reason a fallback list matters more than a favourite model.
What changed on 3 November
Two changes landed in the same daily check. First, the roster shrank: the two Poolside models, Laguna S 2.1 and Laguna XS 2.1, were listed as dropped in the previous seven days. A separate free-tier tracker likewise counted 14 free text models on 3 November, down from 17 on 27 October. Second, a new free model appeared: Crouton Alpha, a stealth model released on OpenRouter on 3 November 2026, listed at $0 for input and output with a 1,048,576-token context and text, image, video, file and audio input.
Crouton Alpha comes with a clear caveat. OpenRouter describes it as developed and operated by an anonymous third-party provider during a preview, and states that prompts and completions may be retained by the provider but are not used for training. No end date is stated, and the tracker notes that earlier stealth models lost their free price once their preview ended. Treat it as a temporary free endpoint for experiments, not as a permanent foundation — and do not send it anything sensitive.
Which free models answered on 3 November
Of the 14 models listed in the 3 November check, these 7 answered: Cohere North Mini Code, NVIDIA Nemotron 3 Nano Omni, NVIDIA Nemotron 3.5 Super VL, NVIDIA Nemotron 3 Ultra, NVIDIA Nemotron 3.5 Lightning, Dots Studio Dots3-Note Preview, and XingChen Xing4.0 29B A4B. The free router, openrouter/free, also answered that day and routed to Nemotron 3 Nano Omni.
The other listed models did not all fail for the same reason, which is why the headline count is misleading on its own. Both Google Gemma 4 models returned upstream rate-limit errors, the Thinking Machines Inkling models returned errors saying they are only available inside agent tools, and Recursive Flint returned a server error that day. A model can be listed as free and still be unusable for a plain API call at the moment you need it.
How to use the free tier without getting stuck
Start with the router, then add fallbacks. Call openrouter/free first and let OpenRouter pick an available free model, and keep two or three :free models from different providers as fallbacks so one provider's capacity problem does not stop your app.
Leave headroom for reasoning. Several free models reason before answering, so a short reply can still consume a large token budget. Set a generous maximum-token limit or the reasoning may use up the budget before any answer is written.
Check the live list before you build. The roster changes month to month, and older guides go stale quietly. OpenRouter's own model list, filtered to free models, and its models API are the authority — re-check them on the day you start a project, not the day you read a blog post.
Keep sensitive prompts elsewhere. Free endpoints, and stealth previews in particular, are the wrong place for secrets, private code, or personal data. Use them for prototypes and experiments, and move anything real to a paid endpoint with terms you have actually read.
FAQ
Is OpenRouter really free?
For the :free models, yes at the token level: input and output are listed at $0, with the request limits above. You still need a free account and an API key, and the daily limit is much higher after a one-time $10 credit purchase.
Why did my free model stop working?
Most often it was removed from the free roster, its upstream provider rate-limited it, or it only works inside an agent tool. Check the current free list, try the openrouter/free router, and fall back to a model from a different provider before assuming your key or quota is broken.
Should I build a product on Crouton Alpha?
No, not yet. It is a free stealth preview with no stated end date, an anonymous provider, and a history of earlier stealth models losing their free price when the preview ended. Use it for experiments; build on a model with a stable free or paid listing.
Is this a no-sign-up tool?
No. OpenRouter requires an account and an API key. For chat with no account at all, see the no-sign-up tools in our agents directory instead.