Use OpenRouter free routes with explicit model and availability expectations. Distinguish a named free variant from a router that chooses among eligible free models.
What is free at OpenRouter right now
| Condition | Verified details |
|---|---|
| Current terms | The official model catalog lists models with zero input and output token rates. Availability and free-model request limits can change. |
| Eligible models | inclusionai/ling-3.0-flash-vl:free, nex-agi/nex-n2.5-mini:free, nex-agi/nex-n2.5-pro:free, inclusionai/ling-3.0-flash-sante:free, inclusionai/ling-3.0-flash-fin:free, dots-studio/dots-3-note-preview:free, liquid/lfm-2.5-2.6b:free, nvidia/nemotron-3.5-lightning:free, thinkingmachines/inkling-small:free, poolside/laguna-s-2.1:free, thinkingmachines/inkling:free, poolside/laguna-xs-2.1:free, cohere/north-mini-code:free, nvidia/nemotron-3.5-content-safety:free, nvidia/nemotron-3-ultra-550b-a55b:free, nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free, google/gemma-4-26b-a4b-it:free, google/gemma-4-31b-it:free, google/lyria-3-pro-preview, google/lyria-3-clip-preview, nvidia/nemotron-3-super-120b-a12b:free, openrouter/free |
Last verified · Source ↗
A model can expose a documented :free variant; the separate openrouter/free route selects from available free candidates. Official documentation.
Choose the form that fits the experiment. A named free variant is useful when you need to evaluate a particular catalog candidate. A free router is useful when the exercise permits the selected model to vary. Keep this distinction visible in the result record so a successful routed response is not attributed to a model you did not actually request.
The free router filters for required features and then selects from its eligible pool, returning the model that served the response. Official documentation.
Save that returned identity with the fixture. If the purpose is to compare answer quality, separate results by resolved model. Pooling them under one router label can hide variation and make a later repeat impossible to interpret.
Limits of the free tier
Published limits have not yet been verified for this selection. Check the official limits documentation and your account console.
Last verified · Source ↗
Inspect the live allowance and its qualifying conditions. A free model’s listed inference category does not promise uninterrupted availability or unlimited requests. Keep a queued, unavailable or stopped state in the application rather than assuming a free demonstration can sustain continuous production traffic.
OpenRouter provides separate account preferences for routing to paid and free providers with different data-handling policies. Official documentation.
Use a safe fixture and review the endpoint’s policy before introducing application data. A zero listed inference price does not answer whether a serving provider’s terms meet your requirements. Treat policy suitability as an eligibility condition before comparing output quality.
What ends the free tier
A free route can become unavailable or cease to match the request’s required features. Its identifier and status need the same verification as a paid route. Preserve the chosen route in configuration and inspect the current catalog when a previously passing example stops working.
Decide whether the application should stop or use an explicitly budgeted paid alternative. Do not silently replace a free-only promise with a paid fallback. Keep the fallback set visible in the workload record and ensure every allowed candidate has passed the relevant output and policy checks.
At the end of a prototype, stop unused jobs and retire unneeded access. Even a free-only experiment can keep generating failed traffic or later inherit a changed configuration. Closing the experiment deliberately prevents an old script from becoming an unexplained background dependency.
How to build within the free tier
Design a small evaluation fixture that is useful even when requests are scarce. Include an ordinary example and a difficult case with a verifiable outcome. Inspect both before expanding the sample. This makes each call answer an integration or quality question instead of merely producing another attractive demonstration.
Use a queue with a defined deadline and avoid creating a burst of independent retry loops. When a temporary response includes timing guidance, preserve it in the worker’s recovery policy. If the allowance is exhausted, pause work rather than repeatedly attempting the same task.
Keep requests self-contained enough to compare the result when the free router selects a different candidate. For conversation tasks, preserve the relevant history and expected response contract. Test whether each selected model handles the same structure before assuming that model variation is harmless.
If a task has a downstream side effect, validate the output before applying it. Free access is not a reason to relax the application’s correctness checks. Use a model response to propose a structured action, verify it against the task’s rules and retain a clear failed state when it does not pass.
Alternatives if you need more
Use the free API selection guide to compare documented offers, and inspect Gemini free-tier conditions or Groq access conditions for another evaluation path.
A provider change should preserve the fixture while rechecking the key, endpoint, model, capacity and policy conditions. A familiar client interface can reduce integration work, but it does not transfer an offer from one service to another.
For a paid OpenRouter continuation, compare the full workload categories and configure an appropriate credit boundary. Record the point at which the application should stop if the paid scenario is not accepted.
Use the AI API cost calculator to turn the model and workload you are considering into an estimate.
Last verified · Source ↗
Frequently asked questions
Is openrouter/free a fixed model?
How do I target a particular free candidate?
Does free mean unlimited availability?
Can a paid fallback be added silently?
Why save the resolved model?
Do free providers share one data policy?
Sources
- OpenRouter quickstart ↗
- Models and pricing schema ↗
- Authentication ↗
- Credit and rate limits ↗
- Errors and debugging ↗
- Provider routing ↗
- Free model variants ↗
- Free router ↗
- Workspace budgets ↗
- Prompt caching ↗
- BYOK ↗
- Provider logging ↗
- Batch quickstart ↗
- Latest aliases ↗
Last verified · Source ↗