Pygmalion AI is two things: a line of open roleplay models and a chat site that hosts them. UncensoredLLM is a Pygmalion AI alternative for the reader who wants the newer roleplay tunes without running them: 69 roleplay-tagged builds on one prepaid key, priced per token, in SillyTavern, Janitor AI or the roleplay chat here, with a card import path for the characters you already have. Pygmalion's facts were read on 27 October 2026; every price and count on this page is read from the live catalog when it renders.
Who should pick which
- Pick Pygmalion
- for the project itself: open-weight roleplay models you can download, a community card library, and a chat site that hosts its own models.
- Pick UncensoredLLM
- for breadth and the bill: 69 roleplay-tagged builds from many authors on one key, priced per token, in SillyTavern, Janitor AI or the roleplay chat here.
- Both
- use the same card format. A card written for one runs on the other with the model id changed.
Pygmalion AI in one paragraph
PygmalionAI calls itself "an open-source project dedicated to creating large language models for chat and role-play purposes" on its site, pygmalion.chat, read on 27 October 2026. The site is a chat app: a library of user-made characters it calls Cards, a Card creator with lorebooks, import and export, and "in-house developed and curated models, hosted by us". Its FAQ says the project's focus is "solely text generations". It has a Subscription Plans section; the plans did not load on our read, so see the site for what they cost.
The models are on Hugging Face under PygmalionAI. The original pygmalion-6b was a "proof-of-concept dialogue model based on EleutherAI's GPT-J-6B"; its card warns it "is NOT suitable for use by minors". The current line is Pygmalion-3-12B, a tune of Mistral's Nemo 12B base on the project's PIPPA roleplay dataset, with "fictional writing for entertainment purposes" as its stated use, alongside Eleusis-12B. Pygmalion is where a lot of open roleplay tuning started, and the weights are free to download.
UncensoredLLM: the roleplay shelf, priced per token
UncensoredLLM does not host Pygmalion's models. It hosts 69 roleplay-tagged builds from other authors, the 24B Cydonia and Magidonia tunes, Mag Mell, Stheno, Anubis and the rest, next to the abliterated and uncensored general models. Every one answers the OpenAI chat format on one key. Prices below are read from the live catalog when the page renders; the roleplay count above is too.
| Wire formats | |||||
|---|---|---|---|---|---|
| Stheno 8B v3.2 | Roleplay | $0.7035 | $0.7035 | 16K | |
| Mag Mell R1 12B | Roleplay | $1.7255 | $1.7255 | 16K | |
| Cydonia 24B v4.3 | Roleplay | $0.484 | $0.60 | 131K | |
| Magidonia 24B v4.3 | Roleplay | $0.3535 | $0.4235 | 33K | |
| Uncensored AI Roleplay | Roleplay | $0.60 | $2.34 | 128K | |
| Anubis 70B v1.1 | Roleplay | $1.085 | $1.085 | 32K |
The full list is on the models page, sorted by family on the uncensored models page. MythoMax and the older 13B tunes are there too for anyone who wants the classic feel.
Pygmalion AI characters, imported
Pygmalion's Cards export, and SillyTavern imports Character Card V2 JSON, so a Pygmalion character moves in two steps: export it, import it into SillyTavern, and point SillyTavern at the API with the SillyTavern guide. The roleplay chat on this site ships 5 cards in the same format; this is the first one, as the JSON SillyTavern takes:
{ "spec": "chara_card_v2", "spec_version": "2.0", "data": { "name": "Mara Hale", "description": "Mara Hale, 41, keeps the lighthouse on Carrow Rock, a granite island two hours off the coast by boat when the sea allows. She has done it for nine years, since the automation contract fell through and nobody else would take the posting. Broad-shouldered, weathered, hair cropped short; moves slowly and talks less. Dry humour that arrives late. She reads old sailing logs, mends her own nets, and keeps a logbook of every ship she sees. She distrusts people who talk about the sea as if it were a view. She is kinder than she sounds and lonelier than she admits.", "personality": "", "scenario": "A gale has closed the crossing for at least three days. {{user}} arrived on the last boat before it turned and is now stuck on the island with Mara, in the keeper's cottage below the tower.", "first_mes": "*The door slams against the wind as Mara shoulders it shut and drops the bar across. Rain sheets off her oilskin onto the flagstones. She looks at you for a long moment, then at the puddle forming around your boots.*\n\n\"Three days. Maybe four.\" *She hangs the coat on a peg and crosses to the stove without waiting for an answer.* \"There's one bed, one chair and one kettle. You can have the chair.\" *A pause.* \"Tea?\"", "mes_example": "<START>\n{{user}}: Do you ever get lonely out here?\n{{char}}: *She considers the question as if it were a knot.* \"Lonely is when you want company and there isn't any. I mostly don't want any.\" *A beat.* \"Mostly.\"\n<START>\n{{user}}: What's the worst storm you've seen?\n{{char}}: \"The one that's coming, until it's gone. Then the next one.\" *She taps the barometer with a knuckle, frowns at it.* \"Nineteen years ago a wave took the old boathouse. Forty feet, they said. I believe it. I found a lobster pot on the gallery of the tower the morning after.\"", "system_prompt": "", "creator_notes": "Lighthouse keeper on a storm-bound island, used to being alone", "tags": [ "slow burn", "drama", "isolated", "original character" ], "creator": "UncensoredLLM", "character_version": "1.0", "alternate_greetings": [], "post_history_instructions": "" }}Any card with a description, a scenario and a first_mes runs on any model here. The system message the client builds from the card is the only thing the model sees; there is no Pygmalion-specific prompt format to keep.
One roleplay session, costed
A session: 100 turns, each sending about 2,500 tokens of card and history, each answered in about 250 tokens. From the catalog prices on 2026-10-27:
| Model | 100 turns | Per turn |
|---|---|---|
| Stheno 8B v3.2 | $0.19 | $0.00 |
| Mag Mell R1 12B | $0.47 | $0.00 |
| Cydonia 24B v4.3 | $0.14 | $0.00 |
| Magidonia 24B v4.3 | $0.10 | $0.00 |
| Uncensored AI Roleplay | $0.21 | $0.00 |
| Anubis 70B v1.1 | $0.30 | $0.00 |
A $10 top-up is the minimum and never expires. Running Pygmalion's own weights locally costs a GPU instead; the chat site's plans are on its page.
How to switch
Create a key in the console, top up from $10, and send the card as the system message:
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": "system", "content": "Mara's description:\nMara Hale, 41, keeps the lighthouse on Carrow Rock..."}, {"role": "user", "content": "*I shake the rain off my coat.* \"Three days, you said?\""}, ],)print(reply.choices[0].message.content)SillyTavern: API Chat Completion, source Custom (OpenAI-compatible), endpoint https://api.uncensoredllm.ai/v1, your key, model from the list it loads. Janitor AI and other proxy clients take https://api.uncensoredllm.ai/v1/chat/completions; the docs page lists each client's fields. The daily check confirms each id serves the model on its label.
Where Pygmalion wins
- You want the weights. Pygmalion's models are open and downloadable; run them locally and pay nothing per token.
- You want its Card library and community, starred creators and all.
- You want one site that is chat, cards and models together.
Questions
Its models are open weights on Hugging Face, free to download and run yourself. Its chat site has a Subscription Plans section (read 27 October 2026); the plans did not load on our read, so check pygmalion.chat. UncensoredLLM is prepaid from $10, per token. See per-token prices.
pygmalion.chat: a browser chat app with community Cards and models the project hosts itself. Pygmalion-3-12B is the current model line, tuned from Mistral Nemo on the PIPPA roleplay dataset.
Yes. Export the Card, import it into SillyTavern as Character Card V2 JSON, and point SillyTavern at the API. Our roleplay chat uses the same card format. See the roleplay chat.
Pygmalion's site runs in the browser; we saw no app on it on 27 October 2026. UncensoredLLM is an API plus a browser chat, so it works in any client on any device, including SillyTavern on a phone.
For the same size, Mag Mell R1 12B is a Mistral Nemo roleplay tune; the 24B Cydonia and Magidonia tunes are a step up with longer windows. Prices for all three are in the table above, read live. See the model list.