OpenRouter alternative for Janitor AI: one key, no 5.5% top-up fee, uncensored DeepSeek and GLM without a prefill

An OpenRouter alternative for Janitor AI has to do three things: take the same three proxy fields, run the models Janitor users already route, and answer the character card. OpenRouter does the first two well. On 2026-10-19 its public list held 471 models, its free models gave 50 requests a day, and its Janitor AI app page showed 998B tokens across 350 models. What it did not list that day was a single abliterated build, so the stock DeepSeek and GLM that Janitor users route there carry whatever prefill the chat needs to keep them answering.

This is the Janitor-shaped version of the OpenRouter alternatives page: what a $20 top-up buys on each side, the three proxy fields filled in, the models Janitor users route on OpenRouter next to the abliterated DeepSeek V4 Pro, DeepSeek V4.1 Flash, GLM 5.3 and GLM 5.3 Flash builds on our uncensored LLM API, and what 100 messages cost on each. Every OpenRouter figure carries the date it was read. Every price of ours comes from the live catalog.

Stay on OpenRouter if you want its free models, the largest list, or a roleplay finetune only it serves. Switch to UplinkSwitch if you want the top-up to buy tokens and nothing else, and DeepSeek and GLM builds that answer a character card without a prefill: 7 builds from $0.0549 per 100 messages, prepaid from $10, with no free tier.

OpenRouter is the default proxy in most Janitor tutorials for good reasons. Its public model list held 471 models at 04:49 UTC on 2026-10-19, more than any other gateway we have read. Its FAQ (read 2026-10-19) gives free models 50 requests a day, and 1,000 a day once you have bought at least $10 of credits, which is how a new Janitor account chats for nothing. A provider object in each request can order the providers behind a model or sort them by price, throughput or latency (openrouter.ai/docs/features/provider-routing, read 2026-10-19).

It also publishes what Janitor users route. Its Janitor AI app page (read 2026-10-19) says Janitor AI calls models through the OpenRouter API, and ranks those models by token volume: 998B tokens in total across 350 models, with this month's top of the list below. 6 of the 7 are on our key under the same name, so the last column prices 100 Janitor messages on each at today's catalog rate.

Two things stand out in that list. Every model on it is a stock build, and six of the seven are DeepSeek or GLM. The same list had 1 model named Uncensored and 0 abliterated builds among its 471 entries on 2026-10-19 (our scan, in the OpenRouter uncensored models post). A stock DeepSeek or GLM answers most cards, and when it does not, the fix Janitor users pass around is a prefill: text sent ahead of every message to talk the model out of refusing. The walkthrough that ranked for janitor ai proxy on 2026-10-19 pairs stock GLM 5.3 Flash with one.

OpenRouter's FAQ (read 2026-10-19) says it passes each provider's price through without markup and charges a 5.5% fee, $0.80 minimum, when you purchase credits. On $20 the fee is $1.10, so $20 of credit is a $21.10 charge. Here a $20 top-up is $20, and every cent of it buys tokens.

The fee is the smaller difference. The larger one is the rate per token. OpenRouter bills the stock DeepSeek ids at DeepSeek's list price (read 2026-10-19); the table takes the developer's list price as OpenRouter's rate for both stock models and puts our rate next to it. One message is 4,000 tokens in for the card, persona and recent chat, and 300 out.

On GLM 5.3 Flash, the stock model in the Janitor walkthrough, $20 buys 26,666 messages at list and 49,382 here, with no prefill in either count. The next table adds the builds that need none.

Open any chat, click the provider button at the top (it reads using janitor until you switch), pick Proxy under Provider, then + New under Proxy configurations. Only three fields matter:

Name        UplinkSwitch
Proxy URL   https://api.uplinkswitch.com/v1/chat/completions
API key     sk-uplink-live-...
Model       deepseek-v4-pro-uncensored

The Proxy URL ends in /chat/completions; Janitor wants the full address, not the base URL. The key comes from the console. Model takes any chat model id on the key. Click Add, then Save, and refresh the page: Janitor only picks up a new proxy URL after a refresh, and the header then reads using proxy. The Janitor AI proxy guide has the same setup with the DeepSeek ids, a long-chat cost table and the steps to give Janitor its own key with a monthly limit.

7 builds answer the configuration above. The 6 abliterated and uncensored ones are the DeepSeek and GLM that Janitor users already route, changed so the card is answered as written: an abliterated model has the refusal direction removed from its weights, and an uncensored build was changed to answer instead of refuse. Roleplay Uncensored is an independent model for character chat. None of them takes a prefill, so nothing rides along with the 4,000 tokens Janitor already sends per message. What a prefill costs on every message is worked in the GLM jailbreak vs an uncensored GLM post.

Once a UTC day, every model on the key gets the same fixed prompt, and the input token count works as a tokenizer fingerprint: a swapped model or an added hidden instruction shows up as a changed count. At the last check (Oct 22 UTC), 6 of the 7 builds it has reached were stable. The verdicts are public on the real-model check.

GLM-5.3 Flash Uncensored is the direct answer to the walkthrough's stock GLM 5.3 Flash plus prefill: $0.0549 per 100 messages against $0.0405 for the stock model here, and no prefill tokens on either count. For DeepSeek, DeepSeek V4.1 Flash Abliterated is the Flash-sized build at $0.4025, and DeepSeek V4 Pro Uncensored the Pro-sized one at $0.46.

Three of the builds were set up and run end to end in Janitor on 2026-10-18, and the notes are from that run.

• Roleplay Uncensored. Answers straight away, with no thinking step.
• DeepSeek V4 Pro Uncensored. Thinks briefly before it answers.
• GLM-5.3 Abliterated. Thinks before it answers. If you set Max tokens, give it 1000 or more.

A model that thinks before it answers spends part of Max tokens on the thinking. At Janitor's low defaults it can run out before the reply starts, and the reply stops mid-sentence. Set Max tokens to 1000 or more for GLM-5.3 Abliterated and DeepSeek V4 Pro Uncensored; Roleplay Uncensored has no thinking step and replies at once. Which builds think, and how three of them behaved in SillyTavern, is in the best AI for roleplay.

• Free models. 50 requests a day on its free models, and 1,000 a day after $10 of credits (FAQ, read 2026-10-19). UplinkSwitch has no free tier; the trial is paused and credit is prepaid from $10.
• Catalog size. 471 models on 2026-10-19, with roleplay finetunes such as Euryale, MythoMax and Cydonia that are not on our key. A card or preset written for one of them runs there.
• The app leaderboard. Its Janitor AI app page shows which models Janitor users route and how much, updated monthly (read 2026-10-19). We publish prices and the daily check, not per-app usage.
• Provider routing. Ordering or filtering the providers behind a model, per request. Here each model has one route at one price.

• No fee on the top-up. $20 of credit costs $20, against $21.10 on OpenRouter (FAQ, read 2026-10-19). Each paid top-up locks the rates for 30 days, and billing takes the lowest live rate in that window.
• DeepSeek and GLM without a prefill. DeepSeek V4 Pro Uncensored, DeepSeek V4.1 Flash Abliterated, GLM-5.3 Uncensored, GLM-5.3 Abliterated, GLM-5.3 Flash Uncensored and GLM-5.3 Flash Abliterated. OpenRouter listed no abliterated build on 2026-10-19.
• Under list on the stock ids. DeepSeek V4.1 Flash 1017 at $0.045 in against $0.30 list and GLM 5.3 Flash at $0.09 in against $0.15 list, from the live catalog on 2026-10-22.
• A daily real-model check. Every build's verdict is in the table above and on /proof, once a UTC day.
• A key just for Janitor. Narrow it to the models you chat with, set a monthly spend limit, pause it when you stop. The one-to-one page, UplinkSwitch vs OpenRouter, puts every term side by side.

Only the three fields change. OpenRouter's chat completions address is https://openrouter.ai/api/v1/chat/completions; ours is https://api.uplinkswitch.com/v1/chat/completions. Model ids drop OpenRouter's vendor prefix: its deepseek/deepseek-v4-pro-0813 is our deepseek-v4-pro-0813 (its model list, read 2026-10-19), and the uncensored build of the same model is deepseek-v4-pro-uncensored.

# OpenRouter                                        # UplinkSwitch
Proxy URL  https://openrouter.ai/api/v1/chat/completions   https://api.uplinkswitch.com/v1/chat/completions
API key    your OpenRouter key                        sk-uplink-live-...
Model      deepseek/deepseek-v4-pro-0813              deepseek-v4-pro-uncensored

Save, refresh, and send one message. If the header still reads using janitor, the page was not refreshed. If the reply stops mid-sentence on a thinking build, raise Max tokens. If an error code shows instead, 401, 402, 403 and 429 are answered under Questions below, and every response carries an x-uplink-request-id header to look the request up.

What's the best AI to use on Janitor AI? Any chat model id on the key works in Janitor's Model field with Proxy URL https://api.uplinkswitch.com/v1/chat/completions. For a reply with no thinking step, Roleplay Uncensored ($0.138 per 100 messages of 4,000 tokens in and 300 out). For DeepSeek that answers a character card without a prefill, DeepSeek V4 Pro Uncensored ($0.46), which thinks briefly first. For the lowest price per message among the builds here, GLM-5.3 Flash Uncensored at $0.0549. See the best AI for roleplay.

What are the free OpenRouter models available for Janitor AI? OpenRouter's FAQ (read 2026-10-19) limits free models to 50 requests a day, and 1,000 a day once you have bought at least $10 of credits. Its Janitor AI app page ranks the models Janitor routes by token volume, and on 2026-10-19 DeepSeek V3.2, Nemotron 3 Ultra and GLM 5.2 led this month's list. UplinkSwitch has no free tier: credit is prepaid from $10 and every model bills per token from it, so $10 covers about 18,227 messages on GLM-5.3 Flash Uncensored.

What are some cheaper alternatives to OpenRouter? For Janitor the number to compare is the cost per message after the fee on the top-up. OpenRouter charges 5.5% with a $0.80 minimum when you buy credits (its FAQ, read 2026-10-19), so $20 of credit is a $21.10 charge. UplinkSwitch charges nothing on a top-up, prices the stock DeepSeek and GLM ids under the developer's list, and runs the uncensored builds from $0.0549 per 100 messages. Eight gateways ranked by the fee each adds on $100 of usage, dated per row, are on the OpenRouter alternatives page. See OpenRouter alternatives by fee.

What proxy do I use for Janitor AI? Under Provider pick Proxy, add a configuration with Proxy URL https://api.uplinkswitch.com/v1/chat/completions, your key from the console, and a model id such as deepseek-v4-pro-uncensored. Click Add, then Save, and refresh the page before you send a message. See the Janitor AI proxy guide.

What is the best roleplay proxy for Janitor AI? Compare four things: the cost per message, since Janitor resends the chat with every message; whether the model refuses the card; whether it thinks before it answers, which needs room in Max tokens; and whether one key reaches every model you want to try. The tables above give all four for 7 builds, and the daily check on /proof says whether each one answered the same way today.

Is Janitor AI better with a proxy? A proxy lets you choose the model and pay per token for it. Janitor still sends the character card, your persona and the recent chat; what changes is which model answers, what a message costs, and whether the reply is refused. With a thinking model, set Max tokens to 1000 or more so the reply is not cut off.

Is there a site better than Janitor AI? This post compares the proxy behind Janitor, not Janitor itself. The same key and the same model ids work in SillyTavern as a Custom (OpenAI-compatible) source and in Chub, so a switch of front end does not mean a new account or a new key.

What does proxy error 403 mean? The key is narrowed to a list of models and the Model field names another one (scope_denied). Add the model to the key in the console, or change the Model field.

Why does Janitor show proxy error 401? The API key is missing, mistyped or revoked. Copy it again from the console, paste it into the configuration and save.

Why does Janitor show proxy error 429? Too many requests in a short time for the key, or the model is busy. Wait a few seconds and send again.

What does proxy error 402 mean? The prepaid balance ran out. Replies work again after a top-up.

Get a key: prepaid from $10, no subscription, 30-day price lock.