If a free OpenRouter model that worked with Janitor AI a few weeks ago suddenly stopped responding, the problem may not be your API key or your proxy setup. OpenRouter’s free-model catalog changes often, and providers can add, remove, or reprice free endpoints without much notice.
We checked OpenRouter’s current free-model listings and separated the models worth considering for Janitor AI from the ones built mainly for coding or business-agent workloads. The important catch: treat the model IDs below as a current snapshot, not a permanent list.
How Many Free Models Does OpenRouter Actually Have?
This is where a lot of guides get sloppy, so let’s be direct about it.
Different trackers report different numbers in the same week. One snapshot counts 14 to 15 zero-cost :free model IDs. Others, pulling from OpenRouter’s own free-model router, count in the low-to-mid twenties. The gap usually comes down to what’s being counted — some trackers exclude multimodal or audio-only free endpoints, others include every regional variant.
The honest takeaway: the free-model count on OpenRouter typically sits somewhere between 15 and 25+ at any given point, and it shifts weekly. Don’t rely on a hardcoded number from any article, including this one. Check the current list on OpenRouter’s free-model collection before you configure a new proxy.
Models Worth Trying Right Now
The table below reflects models multiple independent trackers confirmed as live and free in the first week of August 2026. It is a snapshot, not a complete or permanent list.
| Model | Context (Free Endpoint) | RP Fit | Best Use |
|---|---|---|---|
| nvidia/nemotron-3-ultra-550b-a55b:free | 1M | ⭐⭐⭐ | Long conversations, large context |
| google/gemma-4-31b-it:free | 262K | ⭐⭐⭐⭐ | General roleplay, vision support |
| google/gemma-4-26b-a4b-it:free | 262K | ⭐⭐⭐ | Lighter general roleplay |
| nvidia/nemotron-3-super-120b-a12b:free | 262K on this endpoint | ⭐⭐⭐ | Longer chats, general reasoning |
| inclusionai/ling-3.0-flash:free | 262K | ⭐⭐⭐ | Fast responses, agentic-leaning |
| poolside/laguna-s-2.1:free | 262K | ⭐⭐ | Coding-agent focus, weak for RP |
| cohere/north-mini-code:free | 256K | ⭐⭐ | Code generation, not narrative writing |
*RP Fit is our editorial read based on what each model is designed and marketed for — general chat vs. coding/agent use. It is not a controlled roleplay benchmark. See the methodology note below.
A context-window note: some models show a larger context window on their main model page than on their specific free :free endpoint. Nemotron 3 Super, for example, is documented at up to 1M tokens as a model, but its free endpoint currently lists 262K. Always check the specific :free endpoint page, not just the model’s general spec sheet.
How We Evaluated These Models
We grouped currently free models by intended use case — general chat versus coding/agent-focused — rather than claiming a formal roleplay benchmark. Context limits, provider capacity, and model availability change frequently. These notes reflect OpenRouter’s documentation and catalog at the time of our check, not a fixed or permanent ranking.
Why Free Models Disappear or Stop Working
OpenRouter’s free tier isn’t a fixed list. Providers rotate models in and out based on their own capacity and cost decisions, and endpoints can be delisted with little warning.
Several models that were previously popular free choices for roleplay — including Hermes, Llama, and Qwen variants — may no longer be available as free endpoints as of this check. Availability on these specific models varies by source and by week, so verify the exact model ID on OpenRouter before assuming it still works, rather than relying on any single article, including this one.
This matches a pattern users have reported around Janitor AI: a model works fine for weeks, then a proxy or connection error appears overnight with no changes made on the user’s end.
Before assuming your setup is broken, check the model’s status on OpenRouter directly. If the model itself was pulled or repriced, re-entering your API key won’t fix it. For proxy error codes like 429, 502, or “no response from bot,” our complete Janitor AI error-fix guide covers the diagnostic steps.
Understanding the Daily Message Limit
Free models on OpenRouter aren’t unlimited, and this catches a lot of new users off guard.
- Under $10 in purchased credits: roughly 50 free-model requests per day
- After purchasing at least $10 in OpenRouter credits: the daily allowance rises to roughly 1,000 free-model requests per day
- Rate limit: roughly 20 requests per minute, regardless of tier
- Failed requests still count toward your daily quota — a request that errors out isn’t free to retry indefinitely
This explains the “rate limit exceeded” or proxy error 429 message some users see mid-conversation. It’s the daily cap resetting on a 24-hour cycle, not a bug. If you’re hitting this often, our proxy error 429 fix covers the reset timing in more detail.
Free doesn’t mean unmetered. OpenRouter subsidizes these models, and the cap exists to keep that sustainable.
When Free Isn’t Enough: The Cheap Paid Option
If you’re regularly hitting the daily cap or want more consistent quality, the next step doesn’t have to be an expensive frontier model.
DeepSeek remains a common low-cost recommendation in the Janitor AI community for exactly this reason. Your actual monthly cost depends heavily on message volume, context length, and how long your conversations run — there’s no single number that applies to everyone, so treat any specific dollar figure you see elsewhere with caution.
We’ve published a full setup walkthrough for this: How to Set Up DeepSeek on Janitor AI. It covers account creation, API key setup, and proxy configuration step by step.
How to Connect a Free Model to Janitor AI
The setup process is the same regardless of which model from the table above you pick — only the model ID changes in your proxy configuration.
We’ve already covered the full setup walkthrough, including account creation, API key generation, and proxy URL configuration, in our Janitor AI free proxy and API key guide. Follow that guide first, then swap in whichever free model ID you want to test.
Frequently Asked Questions
How many free models does OpenRouter have right now?
It varies by source and by week — typically somewhere between 15 and 25+ at any given time. The roster rotates as providers add or remove capacity, so check OpenRouter’s live free-model collection rather than relying on a fixed number from any single guide.
Why did my free OpenRouter model disappear from Janitor AI?
OpenRouter periodically delists or reprices free models based on provider capacity. If a model you were using stops responding, check its current status on OpenRouter directly before assuming your Janitor AI configuration is broken.
Are Hermes, Llama, or Qwen models still free on OpenRouter?
Availability on these specific models varies by week and by source. Some trackers report them missing from the current free tier; others show them still listed. Check the exact model ID on OpenRouter before configuring Janitor AI, rather than assuming either way.
How many free messages do I get per day on Janitor AI through OpenRouter?
Roughly 50 messages per day with under $10 in purchased credits, rising to roughly 1,000 per day after a $10 credit purchase. A roughly 20-requests-per-minute limit applies either way, and failed requests count toward your daily total.
What’s the best free model for uncensored roleplay right now?
There isn’t a confirmed winner. Free-model availability and provider-side filtering change frequently, and OpenRouter’s current lineup isn’t specifically built for Janitor AI roleplay. Treat any specific model recommendation, including ours, as a starting point to test rather than a guarantee.
Do I need to pay to use Janitor AI?
No. Janitor AI itself is free. You only pay if you choose a paid model provider like DeepSeek or Claude instead of a free OpenRouter model or the built-in JanitorLLM.
