Mistral is the European model provider that Janitor AI users overlook, and it is one of the easiest to connect: its API speaks the OpenAI chat-completions format natively, so it goes straight into Janitor's proxy field with no compatibility layer, no third-party relay and no :free tag to chase. Per token it is not free, but it is cheap enough that the price is worth stating up front, and its pricing page lists a free plan with a monthly allowance of API credits that, if it applies to you, covers a lot of roleplay.
This page gives the direct values, the arithmetic, the OpenRouter alternative with today's ids and prices, and the honest list of things no proxy changes. We run a proxy network of the IP-changing kind, and the last section says where that fits, which is almost nowhere on this topic.
The direct route: Mistral's own API
Mistral's API reference documents the chat endpoint at https://api.mistral.ai/v1/chat/completions, authenticated with an Authorization: Bearer header, and its own example uses the model mistral-large-latest. That is exactly the shape Janitor's proxy field expects.
| Field | Value |
|---|---|
| Config Name | Mistral direct |
| Proxy URL | https://api.mistral.ai/v1/chat/completions |
| API Key | a key from Mistral's console (La Plateforme) |
| Model | mistral-large-latest, or a smaller model name from Mistral's model list |
Open a chat, tap the menu, choose API Settings, then Proxy, then Add Configuration, fill those in, save, refresh, and send a test line. A network error at this point is almost always the URL missing /chat/completions, and a 401 is the key; our network error and 401 guide has the curl test that tells the two apart.
What it costs, at listed prices
Mistral's pricing page, read on 24 August 2026, listed Mistral Large at $0.50 per million input tokens and $1.50 per million output tokens, and described most models as priced per million tokens with input and output counted separately, batch processing at half price, and cached input at up to 90 percent off.
The roleplay arithmetic: a request carrying 20,000 tokens of character, memory and history costs 20,000 at $0.50 per million, which is one cent, plus a 300-token reply at $1.50 per million, a twentieth of a cent. A hundred messages an evening on Mistral Large is roughly a dollar. Compare Claude Sonnet 5 at about four cents a message on the same context, in our Claude guide, or DeepSeek's own API, which lands in the same pennies range as Mistral.
The same pricing page lists a free plan, described around Mistral's Vibe chat product, that includes $10 a month in API credits. At Mistral Large's listed input price, $10 is twenty million input tokens, or a thousand of those 20,000-token messages. We are quoting the page as it read on the day; plan contents change, and whether the credits apply to your account and your region is something to confirm on Mistral's site rather than on ours.
The OpenRouter route
If you already run other models through OpenRouter, Mistral is a model-id change. OpenRouter's public list on 24 August 2026 carried 19 Mistral ids; the useful ones for chat:
| Model id on OpenRouter | Input, per million | Output, per million | Context |
|---|---|---|---|
mistralai/mistral-large-2512 | $0.50 | $1.50 | 262,144 |
mistralai/mistral-medium-3-5 | $1.50 | $7.50 | 262,144 |
mistralai/mistral-small-2603 | $0.15 | $0.60 | 262,144 |
mistralai/ministral-14b-2512 | $0.20 | $0.20 | 262,144 |
Proxy URL https://openrouter.ai/api/v1/chat/completions, your OpenRouter key, one of those ids, and a balance, since none carry the :free tag. OpenRouter blocks requests, free models included, when a balance goes negative, per its rate-limit documentation. The general OpenRouter setup is in our OpenRouter guide.
What no proxy changes
Limits and credits. Through Mistral directly, the rate limit and the credit balance belong to your Mistral account and follow your key; through OpenRouter, they belong to your OpenRouter account. A second key on the same account shares the limit, and a different IP address is not an input, a rule we lay out for every backend in Janitor AI error 429.
Content. What Mistral's models will write is decided by Mistral's terms and the model's training, not by the URL. Refusals come from the provider.
The route's honesty. Both routes above are your own key with an official endpoint. A "free Mistral proxy" from a Discord post is somebody else's key with somebody else reading the chat; Chub's documentation describes such proxies as created by unauthorized third parties who gain access to your data and IP address. With Mistral costing a cent a message, there is nothing to gain from borrowing one.
Where the network kind of proxy fits
Mistral's API is broadly available, and we found no country list restricting it, so the network-proxy case here is narrower than for Claude or Gemini: a school or office network that filters api.mistral.ai or janitorai.com, or Janitor's own regional rules (it blocked UK addresses from 24 July 2025 until age verification replaced the block on 15 June 2026). For that, a static residential or ISP address in a served country, kept for the session, from $0.44 per GB on a usage meter, is the tool. For the price, the credits, the limits and the refusals, it changes nothing, and the honest recommendation on this page is a Mistral key, the direct URL, and a cent a message.
Sources
- Mistral, API reference: the endpoint https://api.mistral.ai/v1/chat/completions, the Bearer header and the mistral-large-latest example.
- Mistral, Pricing, read 24 August 2026: Mistral Large at $0.50 per million input and $1.50 per million output tokens, per-token pricing with batch and cached discounts, and the free plan listing $10 a month in API credits.
- OpenRouter, public models list: the Mistral ids, prices and context lengths, read 24 August 2026; API rate limits.
- Janitor AI Help Center, The Absolute Beginner's Guide to Using a Proxy with Janitor, Part One: the menu path and field names.
- Chub documentation, API connections: the warning about third-party reverse proxies.