WHY MAGIDONIA

Let's talk about why you're here. Why this model. Why tonight.

THE COMMUNITY

There's a Discord server called BeaverAI. It has 9,000+ members. Most of them are here because TheDrummer, one guy from the community, keeps releasing the best roleplay models on the internet.

Not "really good." Not "competitive." Best.

TheDrummer doesn't work for a lab. He doesn't have a PR team or a launch strategy. He posts updates in the Discord. The community tests them. He iterates. New version drops on HuggingFace. People download it. Word spreads.

This is how open-source actually works when it works well.

THE CYDONIA LINE

In 2025, TheDrummer released seven major versions of the Cydonia model line. Seven. Each one better than the last.

Here's the architecture: Cydonia-24B-v4.3 uses Mistral-Small-3.2-24B-Instruct as its base. Good. Then there's Magidonia-24B-v4.3 which uses Mistral-Small-3.1-24B-Magistral as its base.

The community has spoken: Magidonia wins. Not by a tiny margin. By a lot.

Why? The Magistral base is more coherent, better at following subtle instructions, and less prone to the weird glitches that even excellent models sometimes have. It's a better foundation for fine-tuning.

EVIDENCE FROM THE FIELD

Don't take our word for it. Here's what the community says about Magidonia-24B-v4.3:

"Best version of Cydonia to date"

"Approaches the level of Skyfall" (TheDrummer's 36B model, which costs 3x the VRAM)

"Takes its own initiatives in the story"

"Swipes are rarely ever needed" (swiping = regenerating to get a good response)

These aren't marketing claims. These are people who've been testing models for months. They know what they're looking at.

THE NUMBERS

Cydonia-24B-v4.3 (the regular version, not Magidonia) pulls approximately 415,000 downloads per month from HuggingFace. Per month.

It's listed in every major SillyTavern preset collection. Every Discord bot framework recommends it. If you're new and ask "which model should I use," someone will say Cydonia within the first response.

Magidonia gets recommended even more often. The community choice.

THE COMPETITION

What's the alternative? Dolphin-Mistral-24B-Venice? Good model. Hearthfire-24B? Solid. PaintedFantasy-24B? Gets the job done.

None of them match Magidonia's consistency. And when you look at community merges - projects like Maginum-Cydoms-24B that combine multiple models together - people are merging different versions of Cydonia with each other. Not replacing it with something else.

That's the signal. When expert users combine models, they reach for Cydonia variants. Not the competition.

CENSORSHIP: UNCENSORED OUT OF THE BOX

Here's what you get with Magidonia: a model that works for roleplay without gymnastics.

One user summarized it perfectly: "I haven't gotten any refusals so far."

The model was fine-tuned for creative fiction. It understands context. It understands that roleplaying a villain doesn't make you a villain. It doesn't second-guess you every third response like some commercial models do.

Is it dangerous? No more than any 24B model. It's not trying to jailbreak itself. It's trying to tell a good story with you.

THE HERETIC VARIANTS

If you want even more freedom, the community maintains heretic variants:

coder3101/Cydonia-24B-v4.3-heretic-v2
coder3101/Cydonia-24B-v4.3-heretic-v4
coder3101/Cydonia-24B-v4.3-heretic-absolute-heresy

These use abliteration techniques to remove residual safety training. They're for users who want maximum freedom and accept the slight tradeoff in coherence.

Start with base Magidonia though. It's already not stopping you. And abliteration can slightly degrade the model's ability to stay coherent over long conversations. The base version is the sweet spot.

THE 12B QUESTION

You have 96GB of RAM. You're asking: why not use Rocinante-X-12B instead? It's smaller, faster.

Rocinante-X-12B is the best 12B model. No argument. But one user who tested both said it plainly: "Can't match the detailed feel of 24B."

There's something about the extra parameters. More layers, more attention heads, more room for the model to build a rich internal picture of the character it's playing. It shows in the output.

With your hardware, there's no reason to compromise. You've got the headroom. Use it.

THE MISTRAL ADVANTAGE

Under the hood, Magidonia is built on Mistral's architecture. That matters.

Mistral models excel at instruction following. They handle chat templates cleanly. They work well with long context windows. The engineering is solid.

You'll see this when we get to parameters. Magidonia plays nice with settings that confuse other models. It stays stable. It doesn't drift. It's the architecture doing what it was built for.

SUMMARY

You're using Magidonia-24B-v4.3 because:

1. The community tested it. Extensively. For months. 2. It's specialized for what you're doing: character-driven creative fiction. 3. It's uncensored enough for roleplay without being weird about it. 4. The architecture (Mistral) is battle-tested and stable. 5. Your hardware can run it without compromise. 6. The fine-tuning is excellent. TheDrummer knows what he's doing.

Let's get to the installation.