OpenRouter Free Models in 2026: Limits, IDs & How to Use
OpenRouter’s :free models charge $0 per token but cap at 20 requests a minute and 50 a day. A one-time credit purchase of at least 10 credits lifts the daily cap to 1,000; the higher ceiling is actually granted from 9 credits to absorb OpenRouter’s 5.5% top-up fee. Each swipe counts as its own request. Rate limits are global per account, so extra API keys do not help.
Free models are OpenRouter’s main draw for Janitor AI and SillyTavern users: a working OpenAI-compatible endpoint with current open-weight models at no per-token cost. The trade-off is request limits and pooled upstream capacity, not quality.
Key takeaways
- Free model IDs end in
:free. Token prices are $0 for input and $0 for output- Hard caps: 20 requests a minute and 50 a day; 1,000 a day after you have bought 10 credits
- Each swipe or retry counts as a request, so a 50-a-day cap is 25 messages at one swipe each
- OpenRouter grants the 1,000-a-day cap from 9 credits to absorb a 5.5% top-up fee (minimum $0.80)
- A negative balance blocks free models too; keep your account slightly above zero
What counts as a “free” OpenRouter model?
A free model is any catalog entry with a :free suffix on its ID. OpenRouter’s docs describe these as a separate variant: they have the same base-model quality as the paid version but their own, slower upstream capacity. You can see the current list at openrouter.ai/models?max_price=0. Free IDs rotate, so prefer copying one from that page over an old guide.
A sample from the current list:
google/gemma-4-26b-a4b-it:freegoogle/gemma-4-31b-it:freenvidia/nemotron-3-ultra-550b-a55b:freenvidia/nemotron-3.5-super-vl:freecohere/north-mini-code:freethinkingmachines/inkling:freeliquid/lfm-2.5-2.6b:free
For roleplay on Janitor AI, the Gemma and larger Nemotron entries are usually the ones that behave best; coding-tuned free models often refuse chat prompts. The right choice shifts month to month, so test two or three before committing a character’s history to one.
What are the OpenRouter free model limits?
OpenRouter publishes a short table on its API-reference limits page:
| Credits purchased (all time) | Requests per minute | Requests per day |
|---|---|---|
| Less than 10 | 20 | 50 |
| At least 10 | 20 | 1,000 |
Three things catch people out:
- “All-time credits purchased”, not current balance. Once you are past 10 credits in lifetime purchases, the 1,000-a-day cap sticks even after you spend them. There is no way to go back.
- The 9-credit grant. Because OpenRouter charges 5.5% on top-ups (minimum $0.80), a $10 purchase lands about $9.45 in credits. The 1,000-a-day cap is actually granted starting at 9 credits, so the nominal $10 top-up still works.
- Rate limits are global per account. Making a second account or more API keys does not help: OpenRouter governs capacity per user.
Swipes count. Every retry, every reroll, every “regenerate” button is its own request. Fifty requests a day at a half-swipe average works out to about 33 messages; at one swipe per message, 25.
How much is a credit top-up really worth?
OpenRouter’s FAQ says: 5.5% fee on credit purchases, $0.80 minimum, crypto 5%. On a $10 top-up:
- Cost: $10.00
- Fee: $0.80 (the minimum beats 5.5% of $10, which is $0.55)
- Credits in your account: about $9.20
That $9.20 only matters if you plan to use paid models. For free-model users, the point of the $10 is the cap unlock, not the balance. You can let the balance sit; it will not expire for a year.
Using OpenRouter free models with Janitor AI
Open a chat on janitorai.com, go to the chat’s API settings, and add a new proxy configuration:
- Proxy URL:
https://openrouter.ai/api/v1/chat/completions - Model name: any current ID ending in
:free(copy from openrouter.ai/models?max_price=0) - API key: your OpenRouter key
Hit the built-in connection test. If it fails:
- 401: key pasted with a space or line break, or revoked. Create a fresh key on OpenRouter and paste again.
- 402: account balance is negative, which blocks free models. Top up to bring it above zero.
- 404: model ID is misspelled, retired, or no longer exists. Copy a current one from the models page.
- 429: daily cap hit (50 or 1,000), per-minute cap hit (20), or the upstream provider is at capacity. If
error.metadata.provider_codeis set, the upstream provider is to blame; retry or switch model.
For the full setup around proxies, chats and billing, see the Janitor AI proxy guide. For the DeepSeek-direct alternative, see how to use DeepSeek on Janitor AI.
Using OpenRouter free models with SillyTavern
SillyTavern’s “Chat Completion” provider list already includes OpenRouter. Set the provider to OpenRouter, paste your API key, and pick a :free model from the dropdown. Two settings to adjust early:
- Context size. Free models often have smaller context windows than the paid flagship version. Check the model page before setting context size above 16k.
- Streaming on. OpenRouter streams responses over SSE; with streaming off, long replies can hit a per-request upstream timeout and arrive as an empty finish.
What will a free vs paid day cost?
The proxy cost calculator handles this live, but a rough shape for a typical 60-message day with an 8k context and half a swipe per message:
- Free
:freemodel: $0, but 90 requests a day is over the 50-a-day cap. You need the 1,000-a-day tier (one-time $10 top-up), or you ration to 33 messages. - DeepSeek Flash paid: about $5 a month off-peak, no caps beyond concurrency.
- Gemini 2.5 Flash paid: about $7.50 a month on current prices (and the paid Flash tier for 3.6 and 3.8 doubles on 1 January 2027).
The free cap is not the only reason to pay. Upstream capacity on free variants is pooled across every OpenRouter user, so 429s cluster at peak hours and some free models stall for ten or fifteen seconds before the first token. A paid model is usually faster, not just unlimited.
When is a paid model cheaper in practice?
If you send more than about 25 messages a day with swipes, the “free” tier breaks and you are either rationing or switching to paid anyway. DeepSeek Flash at off-peak rates (currently $0.15 per million input tokens, $0.60 per million output) stays under $5 a month for most 60-messages-a-day users. On paid models there are no per-day request caps, and OpenRouter’s price is the provider’s: no markup on inference.
Three cases where paid wins even at light use:
- Specific model ID. You want
anthropic/claude-haiku-4.6oropenai/gpt-5.1-nanospecifically. There is no free variant. - Long context. Free models often cap at 8k or 16k context, which blows up on long character histories.
- Latency matters. Live voice or avatar workflows cannot afford pooled-capacity queues.
Sources & references
- OpenRouter API docs, API Credit & Rate Limits: the full 20/min, 50/day, 1,000/day table and the “all-time credits purchased” rule. Checked 2026-11-04. openrouter.ai/docs/api-reference/limits
- OpenRouter FAQ, Pricing and fees: the 5.5% credit-purchase fee, $0.80 minimum, 5% crypto fee, and the 1-year unused-credit expiry. Checked 2026-11-04. openrouter.ai/docs/faq
- OpenRouter models browser filtered to free: the current list of
:freemodel IDs. Checked 2026-11-04. openrouter.ai/models?max_price=0 - DeepSeek API Models & Pricing: current Flash and Pro prices, peak hours, cache hit and cache miss rates. Checked 2026-11-04. api-docs.deepseek.com/quick_start/pricing
What will my proxy cost per message and per month?
Prices checked 2026-11-04 from each provider’s official pricing page (sources below). Estimates assume every swipe resends the full context and exclude tax. A 1,000-token reply is roughly 750 English words. Nothing you enter leaves your browser.
Proxy doctor: what does this error mean?
Paste the error your proxy shows (for example “429”, “failed to fetch” or “model not found”).
All common proxy errors
- 401 Unauthorized / invalid API key
- Cause: The key is wrong, revoked, or pasted with a space or line break. Fix: Create a fresh key in the provider dashboard, paste it again with no spaces, and check it belongs to the same provider as the endpoint URL.
- 402 Payment required / insufficient balance
- Cause: Your balance is empty or negative for a paid model. Fix: Top up the provider balance, or switch to a free model. On OpenRouter, a top-up of 10 credits also lifts the free-model cap from 50 to 1,000 a day.
- 403 Forbidden
- Cause: The key does not have access to this model, the account is restricted, or the model is unavailable in your region. Fix: Check the model is enabled for your account and region in the provider dashboard, or pick another model.
- 404 Not found / model not found
- Cause: The model ID is misspelled, retired, or the endpoint URL is wrong. Fix: Copy the exact model ID from the provider’s model list (IDs are case-sensitive), and make sure the URL ends in /chat/completions.
- 429 Too many requests / rate limited
- Cause: You hit a per-minute or per-day cap. OpenRouter free models allow 20 a minute and 50 a day (1,000 a day after buying 10 credits). Busy upstream providers also return 429. Fix: Wait and retry, slow down swipes, try another free model, or add credit. Each swipe counts as a request.
- 5xx / overloaded / timeout
- Cause: The upstream provider is down, overloaded or slow; it is not your settings. Fix: Retry in a few minutes or switch provider. Large contexts make timeouts more likely, so trimming chat memory helps.
- “Failed to fetch” / network error
- Cause: The request never reached the provider: usually a wrong endpoint URL, a typo in https, or a browser extension or network blocking it. Fix: Re-check the full endpoint URL (for example https://openrouter.ai/api/v1/chat/completions), disable blockers for the site, and try another network.
- Context length exceeded
- Cause: Character card + persona + memory + chat history is larger than the model or your settings allow. Fix: Lower the context size, shorten the character card or lorebook, or start a new chat with a summary.
- Empty or cut-off replies
- Cause: The reply-length (max tokens) setting is too low, or a reasoning model spent its tokens thinking. Fix: Raise max new tokens, or turn reasoning off for chat models that support it.
- Refusals / content blocked
- Cause: The model or provider’s content policy declined the request. This is policy, not a bug. Fix: Pick a provider whose published usage policy fits what you use it for. We don’t publish ways around provider safety rules.
Frequently asked questions
Are OpenRouter free models really free?
Yes. Models whose IDs end in :free charge $0 per input and $0 per output token. OpenRouter caps them at 20 requests a minute and 50 a day; buying at least 10 credits once lifts the daily cap to 1,000. Each swipe or retry counts as its own request.
What are the OpenRouter free model limits?
20 requests a minute, 50 a day under 10 all-time credits purchased, 1,000 a day at or above 10. OpenRouter grants the higher ceiling from 9 credits to absorb its 5.5% top-up fee. Limits are global per account; extra API keys do not help.
How do I unlock 1,000 free requests a day on OpenRouter?
Buy credits once on the Credits page. A $10 top-up lands about $9.45 of credits after the 5.5% fee ($0.80 minimum). The higher daily cap sticks even after the credits are spent, because it is keyed to all-time credits purchased, not your current balance.
Can I use OpenRouter free models on Janitor AI or SillyTavern?
Yes. Paste https://openrouter.ai/api/v1/chat/completions as the proxy URL, your OpenRouter key, and a model ID ending in :free. Each swipe counts as a request, so the 50-a-day cap goes fast; raise the cap by buying 10 credits, or switch to a paid model.
Why do OpenRouter free models still return 402 errors?
A negative account balance blocks free models too. If a paid test request dipped your balance below zero, top up to bring it back above zero. Free-model requests themselves do not drain credits.
Do OpenRouter free models have usage limits beyond rate limits?
Only on free variants with daily request caps and per-minute caps. There is no per-token cap on paid variants beyond what you can afford. Free models also pool capacity across all users, so 429s during peak hours are normal.