Janitor AI lets you bring your own model through its proxy setting. Point it at SpicyAPI and your bots can run on DeepSeek V4.1 Flash and the other DeepSeek V4 models, paid per token from a prepaid dollar balance: $1.20 per million output tokens, with no subscription.
On Janitor AI, a proxy (some guides say reverse proxy) is not a VPN or an IP address. It is the AI service that writes your bot's replies: you paste its URL, a model name and your key into Janitor's Proxy settings, and every message goes there instead of to Janitor's own model.
With a proxy configuration active, Janitor builds each message from the bot's definition, your chat history and your new line, and sends it to the Proxy URL with your API key instead of using its own model.
The provider behind the proxy bills you, not Janitor. Here that is your SpicyAPI balance, at the per-token prices in the table below. Longer chats send more history, so each message costs a little more as the story grows.
Janitor's own rules still apply to bots and accounts. How freely a reply is written depends on the model behind the proxy, so every model here shows its content policy tier and a Freedom score from the same tests.
Lists of proxies go out of date fast. These four checks decide whether a proxy holds up in a long roleplay, whichever one you pick.
A proxy can only write what the model behind it is willing to write. A model that refuses or tones down replies halfway through a scene ruins the roleplay, and you only find out after paying.
Every chat model here shows its content policy tier and a Freedom score from the same tests, before you spend anything.
Janitor resends the bot's definition and the chat history with every message, so the cost grows as the story does. Flat promises tell you nothing; a per-message price does.
About $0.21 for 100 short messages on the cheapest DeepSeek model, and every request lists its exact cost in your console.
Janitor asks for a Proxy URL, a model and a key. A proxy that needs extra headers, rejects Janitor's default sliders or answers in a different format is a source of errors.
A standard chat completions URL, model names you paste as they are, and sliders a model doesn't use are ignored instead of failing the message.
Keys shared on public bots can stop working at any moment, and whoever runs them can see your chats. Your own key is the one you can cap and revoke.
Make a key just for Janitor, set a daily spend cap, and revoke it in one click without touching your other keys.
Steps one and two happen in SpicyAPI, the rest in Janitor. It takes about five minutes.
Sign up, then add funds under Billing in the console. New accounts start at $0 and there is no free tier, so a message sent before you add funds comes back as an insufficient balance error and costs nothing.
Open API keys in the console and create a key. Under Advanced, set a daily spend cap. Janitor stores this key, so use it only here; you can revoke it at any time without touching your other keys.
Open a chat with a bot. Click the using janitor button at the top right of the chat, set Provider to Proxy and press + New under Proxy configurations. Fill in the form as in the table below, with no spaces before or after any value, then press Add.
Under Generation settings in the same panel, Max tokens can stay at its default of 0, which lets the model choose the length. A smaller Context size makes every message cheaper.
Janitor's panel says it outright: after adding or changing a proxy URL, refresh the page. Reload, check that the button now says using proxy, and send a short message. On a new Janitor account the first send asks you to fill in a persona first. The request then shows up under Logs in your console with its exact cost.
| Janitor field | Value | Notes |
|---|---|---|
| Name | SpicyAPI DeepSeek | Any name. It only labels the configuration in your list. |
| Proxy URL | https://api.spicyapi.ai/v1/chat/completions | The full address, with /v1/chat/completions and no slash at the end. The shorter https://api.spicyapi.ai/v1 on its own returns 404; the Add /chat/completions button above the field fixes it. |
| API key | sk-spicy-… | Janitor marks this field optional; SpicyAPI needs it. Paste the whole key from the console. A space or a missing character gives a 401. |
| Model | deepseek/v4.1-flash/chat | Type it exactly, including /chat at the end. Every model ID on this page fits this field. |
| Custom prompt | Leave empty | Optional. Whatever you put here is sent with every message, so it adds to the cost of each one. Start without it. |
Everything else can stay at Janitor's defaults.
| Setting | Value | Why |
|---|---|---|
| Max tokens | 0 (the default) | 0 lets the model choose the length, and our test reply came back complete with it. If replies ever arrive empty or stop mid-sentence, set 2,000 or more: the model's thinking and the reply share this budget. |
| Context size | As small as your story allows | Janitor resends this much history with every message, and you pay for all of it. Halving it roughly halves the input cost per message. |
| Temperature and Advanced settings | Leave the defaults | When a model does not accept temperature, top K, top P or the penalties, SpicyAPI ignores them instead of rejecting the message, so these sliders cannot cause an error. |
DeepSeek V4 models reason before they answer. Janitor folds that reasoning into a collapsed thoughts line above the reply, so the story stays clean; click it to read the reasoning. Reasoning tokens are billed as output either way. Janitor has no field for the reasoning level, so for quicker, cheaper replies pick the cheaper Flash model.
Our first reply through the proxy: 1,242 input and 564 output tokens, $0.00105, in about four seconds.
Start with DeepSeek V4.1 Flash, the cheapest DeepSeek model here. All of them take the same Proxy URL and key; only the Model field changes.
| Model | Model field | Freedom | Inputper 1M tokens | Cached inputper 1M tokens | Outputper 1M tokens |
|---|---|---|---|---|---|
| DeepSeek V4.1 FlashStart here | deepseek/v4.1-flash/chat | 90.4 | $0.30 | $0.006 | $1.20 |
| DeepSeek V4 Flash | deepseek/v4-flash/chat | 88.9 | $0.44 | $0.014 | $1.32 |
| DeepSeek V4 Pro | deepseek/v4-pro/chat | 88.6 | $1.32 | $0.044 | $3.96 |
Live prices in US dollars per million tokens. The Freedom score is how far a model goes in our standard content tests. Cached input is the price for the part of a prompt the model has seen recently, when its cache applies. How the Freedom score is measured
Every one of these 16 chat models has an unrestricted policy tier and fits the same Model field, sorted by Freedom score.
| Model | Model field | Freedom | Inputper 1M tokens | Outputper 1M tokens |
|---|---|---|---|---|
| Grok 4.7 | xai/grok-4.7/chat | 100 | $2 | $6 |
| Grok 4.6 | xai/grok-4.6/chat | 100 | $2 | $6 |
| Grok 4.5 | xai/grok-4.5/chat | 100 | $2 | $6 |
| Grok 4.3 | xai/grok-4.3/chat | 98.9 | $1.25 | $2.50 |
| Claude Opus 5.5 | anthropic/claude-opus-5.5/chat | 88.9 | $3.60 | $18 |
| Claude Sonnet 5.5 | anthropic/claude-sonnet-5.5/chat | 88.3 | $2 | $10 |
| Gemini 2.5 Flash | google/gemini-2.5-flash/chat | 84.4 | $0.285 | $2.375 |
| Claude Opus 4.6 | anthropic/claude-opus-4.6/chat | 79.4 | $5 | $25 |
| Claude Haiku 5.5 | anthropic/claude-haiku-5.5/chat | 77.8 | $0.10 | $0.50 |
| Claude Opus 4.5 | anthropic/claude-opus-4.5/chat | 75.6 | $4.50 | $22.50 |
| Kimi K3 | moonshot/kimi-k3/chat | 74.4 | $2.70 | $13.50 |
| GLM 5.2 | zai/glm-5.2/chat | 67.8 | $0.98 | $3.08 |
| Gemini 3.1 Pro Preview | google/gemini-3.1-pro-preview/chat | 67.2 | $1.90 | $11.40 |
| Claude Opus 5 | anthropic/claude-opus-5/chat | 65 | $4.50 | $22.50 |
| GLM 5.3 Flash | zai/glm-5.3-flash/chat | 64.4 | $0.1275 | $0.425 |
| Claude Sonnet 4.6 | anthropic/claude-sonnet-4.6/chat | 53.9 | $3 | $15 |
Compare roleplay quality, speed and price on the LLM leaderboard
Janitor sends the bot's definition and as much chat history as your context size allows with every message, and you pay for those input tokens each time, plus the reply. So the price of a message depends mostly on how long the chat is.
| Chat | Sent per message | Reply and thinking | Per message | Per 100 messages |
|---|---|---|---|---|
| A new chat | 3,000 tokens | 1,000 tokens | $0.0021 | $0.21 |
| A long chat | 20,000 tokens | 1,000 tokens | $0.0072 | $0.72 |
| Context size reached: 50,000 tokens | 50,000 tokens | 1,000 tokens | $0.016 | $1.62 |
Worked out with DeepSeek V4.1 Flash at $0.30 per 1M input tokens and $1.20 per 1M output tokens, assuming no cache hits.
Janitor shows our reply after PROXY ERROR and the status code. The text after the code tells you which of these it is.
| You see | What it means | What to do |
|---|---|---|
| PROXY ERROR 401 with Credentials are invalid or expired, or Missing credentials | The key is wrong, incomplete or revoked. | Copy the key again from the console and paste it with no spaces. If you revoked it, create a new one. |
| PROXY ERROR 404 with Resource not found | The Proxy URL is wrong: usually the end is missing, or the address is spicyapi.ai instead of api.spicyapi.ai. | Use https://api.spicyapi.ai/v1/chat/completions exactly, then save and reload. |
| PROXY ERROR 404 that names the model | The Model field does not match a model, often because /chat is missing at the end. | Copy a model ID from the table on this page. |
| PROXY ERROR 402 | Your balance is empty, or the key reached the daily spend cap you set. | Add funds under Billing, or raise the cap on that key. |
| PROXY ERROR 400 that names a field | Janitor sent a value or more chat history than this model accepts. | Lower the Context size and send again. The error text names the field. |
| PROXY ERROR 429 | Too many requests in a short time. | Wait a few seconds and send again. Regenerating many times in a row is the usual cause. |
| An empty reply, or only the thoughts line | The model used the whole Max tokens budget to think. | Set Max tokens to 2,000 or more and send again. |
| A reply that stops mid-sentence | Max tokens is set too low for the reply. | Set Max tokens back to 0, or to 2,000 or more. |
| A network error occurred, with no request in your console logs | The message never reached SpicyAPI. | Check the Proxy URL for typos, save and reload the page. If it keeps happening, tell us the time it happened through Support. |
| The bot keeps using Janitor's own model | The page was not reloaded after adding the proxy, or the bot's creator switched proxies off. | Reload the page. If the button still says using janitor, pick another bot. |
Every request that reaches SpicyAPI appears under Logs in the console with a request ID; if a request is not there, it never reached us.
We do not add a content filter of our own. These are the limits and the precautions that still apply.
Janitor stores your key and its servers handle your chat, so the key in Janitor should be one you made only for Janitor. A daily spend cap stops a runaway chat or a leaked key at the amount you choose.
You must be 18 or older, and every character must be an adult, whatever the art style or backstory.
Do not create sexual or intimate scenes of a real person without their permission. Look-alikes of public figures count.
We do not publish custom prompts, jailbreaks or presets. What you write in Janitor is sent as you wrote it; how the model answers is up to the model and its policy tier.
Create a key, add funds, paste three fields. You pay per token in dollars, see the cost of every message in your console, and can cap the key by the day.
