Free DeepSeek Proxy for Janitor AI: Setup and Fixing 429 Errors

DeepSeek is no longer free for Janitor AI: DeepSeek's own API charges per token, and no DeepSeek model on OpenRouter is free. Direct from DeepSeek, deepseek-flash costs $0.15 to $0.30 per million input tokens (DeepSeek's price list, November 7, 2026). If it must cost nothing, use Janitor's built-in model, JLLM, or one of OpenRouter's 12 free models.

Janitor's own quickstart explains why so many setups break. Last updated on August 1, 2025, it calls OpenRouter the "easiest free option" and tells you to paste deepseek/deepseek-chat-v3-0324:free. On November 7, 2026, OpenRouter listed that model only as paid: $0.29 per million input tokens and $1.14 per million output. The values below work today, with the prices behind them and a fix for each error, 429 first.

Is DeepSeek still free on Janitor AI?

No route through DeepSeek or OpenRouter is free in November 2026. We read OpenRouter's model list on November 7, 2026: 460 models, 19 of them from DeepSeek. None of the 19 carried the :free ending that marks a free model. Our count of September 28, 2026 found 16 DeepSeek models and none free. Free models from other makers fell from 17 to 12 between the two dates.

DeepSeek's own API bills every request from your balance. A key with nothing behind it gets error 402, which DeepSeek's error page explains as "You have run out of balance." The price list also mentions a "granted balance", spent before a top-up when an account has one. It promises none to new accounts.

Three things in this setup still cost nothing:

• JLLM, Janitor's default model. Janitor's help center calls it "free, built-in, and ready to go out of the box". It answers whenever no proxy is configured.
• OpenRouter's free models. Twelve ids ended in :free on November 7, 2026, from makers such as NVIDIA, Google and Liquid. They allow 20 requests a minute and 50 a day, or 1,000 a day once you have bought 10 credits.
• openrouter/free, which OpenRouter describes as "a router that selects free models at random". The model behind a reply can therefore change.

Hosts that resell model access sometimes add free daily quotas. MegaNova publishes Janitor setup guides of its own. Its pricing page advertised 550 free messages a day on "eligible models" in November 2026, and it does not say whether DeepSeek is one of them. It lists DeepSeek V4.1 Flash at $0.10 per million input tokens and $0.30 per million output tokens. Any such host sits between your chat and the model, so it receives every message you send.

Is a Janitor AI proxy a network proxy?

On Janitor AI, "proxy" means the connection to an outside language model. Janitor's glossary puts it plainly: "The proxy acts as the middleman between your chat and the model provider." The proxy decides which model writes the replies. Your IP address and your location stay the same.

A network proxy, such as a residential or datacenter IP, routes your traffic through another address. It connects Janitor to no model. A new address also leaves a quota error in place, because both providers count limits per account. Our roundup of the best free Janitor AI proxy setups covers the other model backends people connect.

Which fields does Janitor AI need?

Janitor's quickstart lists five fields in a proxy configuration: Config Name, Model Name, Proxy URL, API Key and Custom Prompt. Three of them decide whether the connection works:

• Proxy URL: the full address of the provider's chat endpoint, ending in /chat/completions.
• API Key: the secret key from the same provider.
• Model Name: the exact model id, copied character for character.

The config name and the custom prompt are yours to choose. Key, URL and model must all come from one provider. A DeepSeek key sent to OpenRouter's address is an invalid key there, and OpenRouter answers 401.

How do you run DeepSeek in Janitor AI?

Both routes reach the same DeepSeek model at the same price in November 2026. OpenRouter suits you if you already hold a key from Janitor's quickstart. DeepSeek direct takes one company out of the path.

Option A: DeepSeek through OpenRouter (used to be free, now paid)

OpenRouter puts 460 models behind one address and one key. Its free DeepSeek ids, the ones older guides and Janitor's quickstart still quote, are gone. The id to use now is deepseek/deepseek-v4.1-flash-1017. OpenRouter's list describes it as "served by DeepSeek under the deepseek-flash API model name". It costs what DeepSeek charges: $0.15 per million input tokens and $0.60 per million output tokens off-peak, and twice that in DeepSeek's peak hours.

Steps for the OpenRouter route:

1. Sign up at openrouter.ai.
2. On the Keys page, create a key and copy it.
3. Add credit. DeepSeek is paid here, and 10 credits also lift the daily cap on free models to 1,000.
4. In a Janitor chat, open API Settings, choose the proxy option and add a configuration.
5. Paste the three values below, save, refresh the page and send a test message.

Proxy URL:   https://openrouter.ai/api/v1/chat/completions
API Key:     your OpenRouter key
Model Name:  deepseek/deepseek-v4.1-flash-1017

OpenRouter adds a DeepSeek id each time DeepSeek ships a snapshot: 19 were listed on November 7, 2026, against 16 six weeks earlier. If a name stops answering, open OpenRouter's models page, filter for DeepSeek and copy the newest id. The deepseek/ prefix stays; the version after it moves.

Option B: DeepSeek's own API (direct, with no free-tier cap)

Steps for the direct route:

1. Create an account at platform.deepseek.com.
2. Top up a small balance. Requests fail with 402 until there is one.
3. Create an API key and copy it.
4. Paste the three values below into Janitor, save and test.

Proxy URL:   https://api.deepseek.com/chat/completions
API Key:     your DeepSeek key
Model Name:  deepseek-flash

Use deepseek-flash. DeepSeek's first-call page says the name "is now served by DeepSeek-V4.1-Flash-1017", the snapshot of October 17, 2026, at unchanged prices. Older names still circulate in guides:

• deepseek-chat and deepseek-reasoner were discontinued on July 24, 2026, three months after DeepSeek announced it.
• deepseek-v4-flash is a legacy name. DeepSeek still accepts it and serves the same model at the Flash price.
• deepseek-v4-pro is the larger model, at $0.66 per million input tokens and $1.98 per million output tokens off-peak.

If the address above returns an error in your setup, try the form with /v1 before /chat/completions. On September 28, 2026, both forms answered our keyless test with DeepSeek's key error, 401.

DeepSeek runs deepseek-flash in thinking mode by default: the model writes a chain of reasoning before each reply. The switch is a field in the request, and none of Janitor's five configuration fields sets it. In thinking mode DeepSeek also ignores temperature. Its documentation says the setting "will not trigger an error but will also have no effect". Janitor's temperature setting therefore changes nothing on DeepSeek's API.

What does DeepSeek cost on Janitor AI?

DeepSeek bills the tokens you send and the tokens it writes, and the price halves outside peak hours. Peak runs from 01:00 to 04:00 and from 06:00 to 10:00 UTC, Monday to Friday, except Chinese public holidays. Every other hour is off-peak, weekends included. The split began on August 16, 2026.

Input tokens that hit DeepSeek's cache cost $0.003 per million off-peak and $0.006 at peak. Janitor's glossary gives a rule of thumb for the size of a token: "Roughly 1,000 tokens = 750 words." Through OpenRouter, deepseek/deepseek-v4.1-flash-1017 follows the same price windows.

OpenRouter vs DeepSeek direct, side by side

Both routes cost money now. The choice is one key for many models, or one company fewer between you and DeepSeek.

The prices are per million tokens for deepseek/deepseek-v4.1-flash-1017 and deepseek-flash, from OpenRouter's model list and DeepSeek's price list, read November 7, 2026.

How do you fix the 429 errors?

A 429 is the HTTP status for "Too Many Requests". The standard, RFC 6585, defines it as a client that "has sent too many requests in a given amount of time". The answer may also carry a Retry-After header that says how long to wait. Where the limit sits depends on the route.

On OpenRouter, a 429 has two possible sources, says its limits page. The first is OpenRouter's own cap on free models:

• 20 requests a minute on any :free model. Rapid regenerating can reach it within a minute.
• 50 free requests a day for an account that has bought fewer than 10 credits in total, and 1,000 a day at 10 or more.

The second source is the provider behind the model, when it "is rate limiting or at capacity". A paid id such as DeepSeek's carries "no platform-level request cap", so on it only the second source applies.

On DeepSeek's own API, a 429 means "You are sending requests too quickly, or have exceeded a concurrency or QPS limit." The limit for deepseek-flash is 2,500 requests in progress at once per account. DeepSeek counts it "regardless of which API Key is used".

Five ways to clear a 429, roughly in order of effort:

1. Wait a minute. The per-minute cap resets fast, and a Retry-After header, when present, names the wait.
2. Stop hammering regenerate. Every regenerate is a new request against the same quota.
3. Buy 10 credits on OpenRouter. The daily cap on free models rises from 50 to 1,000, and the credit stays in your balance.
4. Switch to a paid model id. On OpenRouter a paid id carries no platform-level request cap.
5. Move to DeepSeek's own API (Option B). Its only limit is the per-account cap above.

None of these fixes changes your IP address. OpenRouter states that "Making additional accounts or API keys will not affect your rate limits, as we govern capacity globally". DeepSeek counts requests per account. A network proxy therefore cannot clear a quota error.

What do the other errors mean?

Janitor shows its own message when a request fails. Searches for it often quote "A network error occurred, you may be rate limited or having connection issues: failed to fetch". Behind such a message sits one of the provider codes below, each explained in the provider's own words.

• Network error or failed to fetch: check the Proxy URL character by character, ending in /chat/completions, and check that key, URL and model come from one provider. Then save, refresh the page and retry.
• 401: DeepSeek says "Authentication fails due to the wrong API key"; OpenRouter says "Invalid credentials". Paste the key again, with no space at either end.
• 402: the balance is empty. DeepSeek: "You have run out of balance." OpenRouter: "Your account or API key has insufficient credits." Top up.
• 403 on OpenRouter: a guardrail block or a moderation flag stopped the request.
• 408 on OpenRouter: the request timed out. Retry.
• 502 on OpenRouter: "Your chosen model is down or we received an invalid response from it." Retry, or pick another DeepSeek id.
• 503: DeepSeek's server is overloaded, or on OpenRouter no provider meets your routing requirements. Retry after a short wait.
• Region or network blocks: if a provider does not serve your country, or your school, office or ISP filters it, we give no steps around that.

Can a network proxy help with Janitor AI?

For every error on this page, no. Limits, balances and model names live in your accounts, and both providers count limits per account. A residential or datacenter IP changes the address a website sees and nothing in a DeepSeek or OpenRouter account.

Our lines do fit a network that needs a known exit, such as a company network that sends all traffic through a proxy. Never send an API key you care about through a free proxy of unknown origin; are free proxies safe explains why. Our page on proxies for DeepSeek covers the DeepSeek API behind a company proxy. Janitor AI error 429 sorts the 429 fix by backend. Our residential proxies are pay-as-you-go at $0.18/GB, and the free proxy checker confirms that a line answers.

What this page could not check

We hold no paid DeepSeek or OpenRouter key and bought nothing for this page, so it shows no replies from either route. Each error meaning comes from the provider's own documentation. Janitor publishes no limits for JLLM, and MegaNova does not name the models its free quota covers. Janitor's help pages were last updated on July 31 and August 1, 2025, and its menu labels move between updates. Model names, prices and OpenRouter's free list change monthly, so every dated figure here is due for a check by December 7, 2026.

Sources

• DeepSeek, API Docs: "Your First API Call", "Models & Pricing", "Change Log" (entries to October 17, 2026), "Rate Limit & Isolation", "Error Codes" and "Thinking Mode", read November 7, 2026.
• OpenRouter, documentation: "API Credit & Rate Limits" and "Errors", read November 7, 2026.
• OpenRouter, models API (the full model list), read November 7, 2026; the same count on September 28, 2026.
• Janitor AI Help Center, "TL;DR Quickstart Proxy Instructions", updated August 1, 2025, and "What is JLLM? And what do all those pesky abbreviations mean?", updated July 31, 2025.
• IETF, RFC 6585, "Additional HTTP Status Codes", section 4, April 2012.
• MegaNova, pricing page, read November 7, 2026, quoted for its own offer.
• HProxy, keyless test of DeepSeek's two endpoint forms, September 28, 2026.