The Drummer Hugging Face profile is where Cydonia, Anubis, Magidonia and Skyfall come from: roleplay fine-tunes of Mistral and Llama bases with the alignment trained away. This key runs 10 of the profile's releases. This post names 5 of them, quotes what each model card says the tune is for, reads their prices from the live catalog, and shows the daily check verdict next to each one. Credit goes to TheDrummer; the weights are theirs and the cards are linked.
Who The Drummer on Hugging Face is
TheDrummer is a Hugging Face account, display name Drummer, that publishes roleplay and creative-writing fine-tunes of open-weight models. On 27 October 2026 the profile listed 222 models and 4,068 followers, a dataset (AmoralQA-v2), and one outside link, a Patreon. The profile's recent activity also shows repos under the BeaverAI organization.
The families on the profile that day: Cydonia and Magidonia (24B), Skyfall (31B and 36B), Rocinante and Rivermind (12B to 16B), Artemis (31B), Orion (26B), Behemoth (123B and 128B) and Anubis (70B). The model cards share a shape: the base, the chat template, a GGUF link with bartowski's iMatrix quants recommended, and a column of user quotes. Sampler settings are rarely given.
This key runs 10 TheDrummer releases. The rest of this post is about 5 of them: what each card says the tune is for, what it costs per token today, and whether the daily check still finds the model on the label. The others have their own model pages: Anubis 70B v1, Artemis v1.1, Cydonia 24B v2, Rocinante 12B v1.1, UnslopNemo 12B v4.1.
TheDrummer models on this key
Each row is one repo under the TheDrummer account, confirmed on 27 October 2026. Base and template are from the card; the downloads column is the card's "last month" figure that day and will have moved since.
| Model | Base (per the card) | Template | Downloads, 30 days | Daily check |
|---|---|---|---|---|
cydonia-24b-v4.3 | Mistral Small 3.2 24B Instruct 2506 | Mistral v7 Tekken | 2,183 | Steady since Oct 23 |
cydonia-24b-v4.1 | Mistral Small 3.2 24B (tagged mistral) | Mistral v7 Tekken | 3,737 | Steady since Oct 26 |
magidonia-24b-v4.3 | Magistral Small 2509 | Mistral v7 Tekken | 941 | Steady since Oct 28 |
skyfall-36b-v2 | Mistral Small 2501, upscaled to 36B | Mistral v7 Tekken (Alpaca for adventure) | 3,620 | Steady since Oct 23 |
anubis-70b-v1.1 | Llama 3.3 70B Instruct | Llama 3 | 175 | Steady since Oct 28 |
Not on the key: Behemoth, Rocinante, Rivermind, Orion and Skyfall 31B. They are on the profile for download.
What each tune is for
- Cydonia 24B v4.3. The current Cydonia. The card lists "much better roleplay" over v4.1 and says thinking tags work but are not needed.
- Cydonia 24B v4.1. The previous Cydonia, "Cydonia Evolved again" per its card. Prose-first roleplay; one card reviewer calls it "probably the best Cydonia".
- Magidonia 24B v4.3. Cydonia v4.3's sibling on the Magistral reasoning base. The card says the two differ mainly in writing style.
- Skyfall 36B v2. Upscaled with continued training "for creativity and RP". The card recommends Tekken and advises against Metharme.
- Anubis 70B v1.1. Built "for character adherence and unalignment" per its card. The 70B of the set, for long casts and strict personas.
The Cydonia 24B line is the one most people mean by "the drummer cydonia". v4.3 is the current release and the id this site's Cydonia 24B v4.3 page describes in full; v4.1 stays listed because its card's reviewers prefer its prose, and the price differs. Magidonia is the same recipe on Magistral, so a scenario that wants a thinking block before the reply goes there. Skyfall is the mid-size pick with the second-most downloads of the five that month. Anubis is the 70B, built for characters that must not drift.
None of the five cards lists a license of its own; the catalog shows what each base carries. For cydonia-24b-v4.3 that reads: none listed (base: Apache-2.0).
The settings the cards give, and which Drummer build to pick
The cards are short on sampler advice and long on templates. Four of the five name Mistral v7 Tekken; Anubis names Llama 3. On this key the template is applied on the server from the model id, so a client sends plain messages and never a Tekken tag. What does carry over is the thinking advice: Cydonia v4.3 and Magidonia v4.3 both say a <thinking> block works and that no thinking works too. Skyfall's card adds that Alpaca suits choose-your-own-adventure play and that Metharme is not recommended. Mag Mell, a different author's merge that shares the roleplay table here, is one of the few of this site's tunes with tested sampler values on its card; the Drummer cards leave temperature to you. Start at 1.0 and lower it if replies drift.
Which to pick, from the cards and the price rows: Cydonia v4.3 for a current, well-tested 24B; v4.1 when a reviewer's preference for its prose matters more than the price; Magidonia when the scene benefits from a reasoning step before the reply; Skyfall for a bigger model; Anubis when the cast is large and a character must hold its voice over a long chat. The 100-message column below turns each choice into a figure.
Live prices and a 100-message chat
Prices per million tokens, read from the catalog on 2026-11-09. Nothing here is typed by hand; the table changes when the catalog does.
| Wire formats | |||||
|---|---|---|---|---|---|
| Cydonia 24B v4.3 | Roleplay | $0.484 | $0.60 | 131K | |
| Cydonia 24B v4.1 | Roleplay | $1.20 | $2.00 | 131K | |
| Magidonia 24B v4.3 | Roleplay | $0.3535 | $0.4235 | 33K | |
| Skyfall 36B v2 | Roleplay | $1.925 | $2.80 | 33K | |
| Anubis 70B v1.1 | Roleplay | $1.085 | $1.085 | 32K |
| Model | Context window | 100 messages |
|---|---|---|
| Cydonia 24B v4.3 | 131,072 tokens | $0.11 |
| Cydonia 24B v4.1 | 131,072 tokens | $0.30 |
| Magidonia 24B v4.3 | 32,768 tokens | $0.08 |
| Skyfall 36B v2 | 32,768 tokens | $0.47 |
| Anubis 70B v1.1 | 32,000 tokens | $0.25 |
Credit is prepaid from $10 with no subscription, and one balance covers these five and every other build on the models page. The per-token rates for all of them are on the pricing page.
Run them by API
Every build takes the OpenAI chat format at https://api.uncensoredllm.ai/v1. The key is created in the console, starts with sk-ullm- and is shown once. The chat template is applied on the server, so you send messages, not Tekken tags.
curl https://api.uncensoredllm.ai/v1/chat/completions \ -H "Authorization: Bearer sk-ullm-live-…" \ -H "Content-Type: application/json" \ -d '{ "model": "cydonia-24b-v4.3", "messages": [ {"role": "system", "content": "You are Mara, a lighthouse keeper. Stay in character."}, {"role": "user", "content": "The storm has the crossing closed. Can I stay?"} ], "max_tokens": 300 }'from openai import OpenAI client = OpenAI(base_url="https://api.uncensoredllm.ai/v1", api_key="sk-ullm-live-…")reply = client.chat.completions.create( model="cydonia-24b-v4.3", messages=[{"role": "user", "content": "Describe the tavern as the door opens."}], max_tokens=300,)print(reply.choices[0].message.content)In SillyTavern, pick Chat Completion with a custom endpoint and the id from the table; the SillyTavern section of the docs has the fields. Janitor AI and the other front-ends take the same URL and key, listed on the docs page.
Or download them
All five repos are public safetensors in BF16. A 24B in BF16 is about 48 GB; the card's GGUF link is the practical download for a single GPU, with a quant sized to the card.
# the safetensors repo, then a GGUF for llama.cpp or KoboldCppgit lfs installgit clone https://huggingface.co/TheDrummer/Cydonia-24B-v4.3# the card points to a GGUF repo and recommends bartowski's iMatrix quantsA GGUF runs under llama.cpp, KoboldCpp or Ollama with a Modelfile. The KoboldAI Lite guide covers the no-GPU route through the same front-end.
Questions
TheDrummer is a Hugging Face account that releases roleplay and creative-writing fine-tunes: Cydonia, Magidonia, Skyfall, Anubis, Behemoth, Rocinante and others. On 27 October 2026 it listed 222 models. It is a publisher of model weights, not a chat app.
The five in the table: Cydonia 24B v4.3 and v4.1, Magidonia 24B v4.3, Skyfall 36B v2 and Anubis 70B v1.1, each served by API on this key and priced per token. Also on the key: Anubis 70B v1, Artemis v1.1, Cydonia 24B v2, Rocinante 12B v1.1, UnslopNemo 12B v4.1. See Cydonia 24B v4.3.
Cydonia 24B v4.3 is a fine-tune of Mistral Small 3.2 24B Instruct 2506, per its card, with the Mistral v7 Tekken chat template. Magidonia 24B v4.3 is the same tune on Magistral Small 2509.
Rivermind-12B-v1 is a 12B release on the TheDrummer profile on 27 October 2026. It is not on this key; the profile has the weights.
Hugging Face hosts community leaderboards, and the UGI leaderboard scores uncensored builds. This site keeps a monthly board of the builds on this key with the daily check next to each score. See the uncensored LLM leaderboard.
It catches a swapped model or an injected prompt by comparing a fixed test prompt’s token count day to day. Two tunes of one base share a tokenizer, so it cannot tell Cydonia v4.3 from v4.1. The verdict column says what it found today. See how the daily check works.
Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.