DeepSeek is one of the most popular models to plug into Janitor AI, mostly because it's cheap. It's also a setup that goes out of date fast. DeepSeek renamed its API models in 2026, and the free DeepSeek route many guides recommend, including Janitor's own quickstart, no longer appears on OpenRouter. This guide gives the settings that match DeepSeek's documentation on October 7, 2026, explains what you'll actually pay, and lists fixes for the errors people hit most.
| Where | In any chat: gear icon (top right) → API Settings → proxy → Add Configuration[1] |
| Proxy URL | https://api.deepseek.com/chat/completions[2] |
| Model name | deepseek-flash (cheaper) or deepseek-v4-pro[2] |
| API key | Created on the DeepSeek platform after you add balance (prepaid)[3] |
| Cost | deepseek-flash: $0.30 per 1M input tokens and $1.20 per 1M output tokens at peak, half that off-peak[3] |
DeepSeek's docs now say to "use deepseek-flash as the model name." The older names deepseek-v4-flash and deepseek-v4-flash-vision-exp "are still accepted, but the corresponding models have been retired," and those requests are served by DeepSeek-V4.1-Flash.[2] If a guide tells you to use deepseek-chat, deepseek-reasoner or a :free model, it predates these changes.
Three ways to use DeepSeek on Janitor AI
| Route | What you need | Cost | Best for |
|---|---|---|---|
| DeepSeek API direct | DeepSeek platform account, prepaid balance, API key | Lowest per-token price; pay as you go | Most users who want DeepSeek specifically |
| OpenRouter | OpenRouter account, key, credits | OpenRouter's listed price per model | Switching between many models with one key |
| janitor+ Router | janitor+ subscription (web only) | Per-token from an in-app wallet, plus the subscription | Not wanting to manage keys at all |
If you'd rather not pay at all, Janitor's free tier includes "unlimited chats with millions of bots," its standard chat models (JLLM) and a "context limit of around 9k tokens."[4] DeepSeek is an upgrade, not a requirement.
How to set up DeepSeek on Janitor AI (direct API)
- Create a DeepSeek platform account at platform.deepseek.com. This is the developer platform, separate from the DeepSeek chat app.
- Add balance. The API is prepaid: "Your balance will be consumed if you use our paid services," and service can be suspended "if your balance is insufficient."[5] Start small. A few dollars goes a long way (see costs below).
- Create an API key and copy it somewhere safe. Treat it like a password.
- Open Janitor AI, start or open a chat, click the gear icon at the top right, then API Settings.[1]
- Select the proxy option (where shown) and click Add Configuration. Fill in:
- Click Add, then Save Settings, refresh the page and send a test message. Janitor's own quickstart recommends the same save-refresh-test order.[1]
Flash or Pro? Both models support a 1M-token context and a "thinking" mode, which is on by default.[3] Flash costs much less (see below). Pro costs about 3.3 times as much for output and 4.4 times as much for uncached input. Most roleplayers start on Flash and switch for scenes where quality matters more.
What DeepSeek costs on Janitor AI
DeepSeek bills per million tokens, input and output separately, with cheaper "cache hit" input when the start of your prompt repeats (which it often does in a long chat):[3]
| Per 1M tokens | deepseek-flash (peak / off-peak) | deepseek-v4-pro (peak / off-peak) |
|---|---|---|
| Input, cache hit | $0.006 / $0.003 | $0.044 / $0.022 |
| Input, cache miss | $0.30 / $0.15 | $1.32 / $0.66 |
| Output | $1.20 / $0.60 | $3.96 / $1.98 |
In US Eastern time, while daylight saving is in effect, peak is 9 p.m. to midnight and 2 to 6 a.m. on weeknights. Weekends are entirely off-peak.[3]
What a message really costs (our estimate). Janitor sends the bot's definition plus recent chat history with every message, so input tokens add up much faster than your typed words. For a message carrying 6,000 tokens of context with a 300-token reply, all billed at the peak cache-miss rate:
- deepseek-flash: 6,000 × $0.30/1M + 300 × $1.20/1M ≈ $0.0022, or about 450 messages per dollar.
- deepseek-v4-pro: 6,000 × $1.32/1M + 300 × $3.96/1M ≈ $0.0091, or about 110 messages per dollar.
Cache hits and off-peak hours lower these figures. Thinking mode can raise them, since reasoning adds output tokens. These are arithmetic examples, not measured bills. Check your usage page on the DeepSeek platform after a session.
Refunds: DeepSeek reviews refund requests for unspent balance. If approved, it refunds "the remaining unspent amount after deducting necessary costs such as handling fees," in one lump sum. Partial refunds aren't supported, and spent amounts aren't refunded.[5] Top up in small amounts.
Using DeepSeek on Janitor through OpenRouter
OpenRouter lets one key reach many models. In Janitor, the setup is the same as above with different values:[1]
- Proxy URL:
https://openrouter.ai/api/v1/chat/completions - Model name: the OpenRouter model ID, for example
deepseek/deepseek-v4.1-flash, which OpenRouter listed at $0.30 per 1M input tokens and $1.20 per 1M output tokens on October 7, 2026.[6] - API key: your OpenRouter key
Janitor's TL;DR Quickstart (updated January 8, 2025) still tells users to enter deepseek/deepseek-chat-v3-0324:free.[1] On October 7, 2026, OpenRouter's public model list included no DeepSeek model with a :free suffix.[6] Free models that do exist on OpenRouter are limited to 20 requests a minute, and to 50 requests a day unless you've bought at least $10 in credits, which raises the cap to 1,000 a day.[7]
janitor+ Router: DeepSeek without API keys
In mid-2026 Janitor launched janitor+ Router, which "lets you use third-party AI models directly inside JanitorAI, without setting up an external proxy or pasting in API keys from another service."[8] According to Janitor's help center:[8]
- It's for janitor+ subscribers only, and web only for now.
- You pay per token at rates shown before you start a chat, from an in-app wallet.
- Subscribers can claim $5 in launch credits and get "a 10% discount on some of the most popular models."
- Janitor says Router "is not a replacement for JLLM+," which it still recommends for everyday roleplay.
Janitor's help center doesn't list which models Router offers. To verify Check the model list before assuming DeepSeek is included. janitor+ costs $12.99 a month in the US App Store[9] and adds "5x more context," priority access and monthly swipes with Janitor's best models. Cancel on the web via Settings → Billing → Manage Subscription (Stripe), or in your app store. Subscriptions "are generally non-refundable."[4]
Fixing common DeepSeek errors on Janitor AI
| Error | What DeepSeek says it means | Fix |
|---|---|---|
| 401 | "Authentication fails due to the wrong API key" | Re-copy the key with no spaces; make sure it's a DeepSeek key, not an OpenRouter one. |
| 402 | "Insufficient Balance" | Top up on the DeepSeek platform. |
| 400 / 422 | Invalid format or parameters | Check the model name is exactly deepseek-flash or deepseek-v4-pro, and the URL ends in /chat/completions. |
| 429 | "You are sending requests too quickly" | Slow down, and avoid rapid rerolls. |
| 500 / 503 | Server error or "overloaded due to high traffic" | Wait and retry, or try off-peak. |
Two more Janitor-side checks: refresh after saving settings, and make sure the configuration you just added is the one selected. Janitor's help center has a separate OpenRouter Error Guide for OpenRouter-specific messages.[10]
Privacy and terms: what to know before you connect
- Your chats go to DeepSeek. With a proxy, the bot definition and chat history are sent to the model provider you choose. DeepSeek's Open Platform is operated by Hangzhou DeepSeek Artificial Intelligence Co., Ltd., and its terms are governed by the laws of mainland China.[5] DeepSeek says it handles your personal information under its own privacy policy.[5]
- DeepSeek may keep records where it suspects illegal activity, and report to authorities.[5]
- Janitor's app is 18+ and rated that way in the App Store.[9] In the mobile app, NSFW content is off by default for app-store reasons. Turn on "Enable NSFW for App" in Settings on janitorai.com, then restart the app.[11]
- Don't share API keys or put personal details into bots. Anyone with your key can spend your balance.
Comparing Janitor with paid companion apps? See our Kindroid review and Character AI alternatives.
Frequently asked questions
What is the DeepSeek proxy URL for Janitor AI?
DeepSeek's API docs give the OpenAI-format base URL as https://api.deepseek.com and show chat requests going to https://api.deepseek.com/chat/completions. Paste the full chat-completions URL into the Proxy URL field in Janitor's API Settings.
What model name do I use for DeepSeek on Janitor AI?
As of Oct 7, 2026, DeepSeek's docs list two model names: deepseek-flash (served by DeepSeek-V4.1-Flash) and deepseek-v4-pro. The older name deepseek-v4-flash is still accepted but routed to V4.1-Flash. Older guides that use deepseek-chat or deepseek-reasoner are out of date.
Is DeepSeek on Janitor AI free?
Not through DeepSeek directly: its API is prepaid and billed per token. On Oct 7, 2026, OpenRouter's public model list showed no free (":free") DeepSeek models, so Janitor's January 2025 quickstart using deepseek-chat-v3-0324:free no longer works as written. Janitor's own JLLM model remains free.
How much does DeepSeek cost per message on Janitor?
It depends on how much chat history is sent each time. Using DeepSeek's peak prices for deepseek-flash ($0.30 per million input tokens, $1.20 per million output tokens), a message with 6,000 tokens of context and a 300-token reply costs about $0.0022, our estimate. That's roughly 450 such messages per dollar at peak. Off-peak rates are half.
Why do I get a 402 error with DeepSeek on Janitor?
DeepSeek's error-code page says 402 means "Insufficient Balance": your prepaid balance has run out. Top up on the DeepSeek platform. A 401 means the API key is wrong, and a 429 means you're sending requests too quickly.
What is janitor+ Router?
A janitor+ feature, web only, that lets subscribers use third-party models inside Janitor without pasting API keys. You pay per token from an in-app wallet, subscribers get $5 launch credits and a 10% discount on some popular models. Janitor+ was $12.99 a month in the US App Store on Oct 7, 2026.
Why is NSFW disabled in the Janitor AI app?
Janitor's help center says app-store rules make it disable NSFW content by default in new app installs. To match the website, log in at janitorai.com in a browser, go to Settings, turn on "Enable NSFW for App," then restart the app. Janitor's app is rated 18+.
Sources
- Janitor Help: TL;DR Quickstart Proxy Instructions, updated Jan 8, 2025 (read Oct 7, 2026)
- DeepSeek API Docs: Your First API Call, read Oct 7, 2026
- DeepSeek API Docs: Models & Pricing, read Oct 7, 2026
- Janitor Help: FAQ: Subscription - Janitorai+, updated Jun 25, 2026
- DeepSeek Open Platform Terms of Service, effective Apr 29, 2026
- OpenRouter public model list (API), read Oct 7, 2026
- OpenRouter Docs: Limits, read Oct 7, 2026
- Janitor Help: janitor+ Router, last updated Jul 31, 2026
- Janitor: Interactive Stories on the US App Store, read Oct 7, 2026
- Janitor Help: OpenRouter Error Guide
- Janitor Help: FAQ: Mobile App Moderation, updated Feb 7, 2026
Error-code meanings are quoted from DeepSeek API Docs: Error Codes, read Oct 7, 2026.