Michael Limberger

Why Magidonia

Here is why this series uses Magidonia, and why that pick still makes sense for a local character chatbot.

The Community

There is a Discord server called BeaverAI with thousands of members. A large share of that crowd follows TheDrummer, a community builder who keeps releasing strong roleplay models and posting them where people can actually try them.

TheDrummer does not work for a big lab. He posts updates in Discord, the community tests them, he iterates, and a new version lands on HuggingFace. People download it, talk about it, and the good versions spread. That loop is open-source working the way it is supposed to.

The Cydonia Line

In 2025, TheDrummer released several major versions of the Cydonia line, each one aiming to beat the last.

The architecture split matters. Cydonia-24B-v4.3 uses Mistral-Small-3.2-24B-Instruct as its base. Magidonia-24B-v4.3 uses Mistral-Small-3.1-24B-Magistral as its base. In practice, Magidonia was the community favorite: more coherent, better at subtle instructions, and less prone to the odd glitches even strong models sometimes show. That Magistral base is a better foundation for this kind of fine-tune.

Evidence From The Field

Here is the kind of feedback Magidonia-24B-v4.3 got from people who had already been testing models for months:

"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 are not launch-day marketing lines. They are notes from people who know what a bad regenerate loop feels like.

The Numbers

Plain Cydonia-24B-v4.3 (not Magidonia) was pulling on the order of 415,000 downloads a month from HuggingFace when these notes were written. It shows up in SillyTavern preset collections, and if someone new asks which model to try, Cydonia often appears in the first reply. Magidonia got recommended even more often for creative chat.

The Competition

Other 24B options exist. Dolphin-Mistral-24B-Venice is capable. Hearthfire-24B is solid. PaintedFantasy-24B can get the job done. For consistency in character work, Magidonia was the pick most of us stuck with. Even community merges like Maginum-Cydoms-24B tend to combine Cydonia variants with each other rather than swapping the whole line out. That is a useful signal: when experienced users blend models, they still reach for this family.

Censorship: Uncensored Out Of The Box

Magidonia is tuned for creative fiction, so roleplay usually works without a pile of jailbreak gymnastics. One user put it simply: "I haven't gotten any refusals so far." The model understands that playing a villain in a story is not the same as being one, and it does not second-guess you every few turns the way some commercial chatbots do.

Is it more dangerous than any other 24B model? Not really. It is trying to tell a story with you, not invent a new jailbreak hobby.

The Heretic Variants

If you want even fewer safety leftovers, 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 to strip residual safety training. You get more freedom and sometimes a little less long-conversation coherence. Start with base Magidonia. It already stays out of your way for most roleplay, and that is usually the sweet spot.

The 12B Question

If you have plenty of RAM, you might ask why not run Rocinante-X-12B instead. It is smaller and faster, and it is a strong 12B model. People who tried both still say the 24B has a richer, more detailed feel. Extra parameters mean more room for the model to hold a character in its head. If your hardware has the headroom, use it.

The Mistral Advantage

Under the hood Magidonia sits on Mistral's architecture, and that shows up in the day-to-day. Mistral models tend to follow instructions cleanly, handle chat templates without drama, and stay usable with longer context. Later, when we tune parameters, Magidonia usually stays stable with settings that make other models drift.

Summary

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

1. The community tested it for a long time, not one weekend.
2. It is specialized for character-driven creative fiction.
3. It is open enough for roleplay without being weird about it.
4. The Mistral base is stable and well understood.
5. Your hardware can run it without forced compromise.
6. The fine-tuning is strong; TheDrummer knows this niche.

Next up: installation.