---
title: "Best uncensored LLM GGUF models, ranked"
canonical: https://uncensoredllm.ai/blog/best-uncensored-gguf-models
generated: 2026-11-09T15:03:49.613Z
source: https://uncensoredllm.ai/blog/best-uncensored-gguf-models
base_url: https://api.uncensoredllm.ai/v1
---

# Best uncensored LLM GGUF models, ranked, with API prices

Ten uncensored LLMs that exist both as a GGUF on Hugging Face and on one API key: base model, what the tune does, the Q4_K_M size from the card, the live price per million tokens and per 100 messages, and today’s real-model check.

Published November 6, 2026 · prices as of November 9, 2026

The best uncensored LLM GGUF lists rank weights by what fits a graphics card. This one ranks builds that exist both as a GGUF on Hugging Face and as an id on one API key, so each row has a download size and a price per million tokens on the same line. The order is the [leaderboard](https://uncensoredllm.ai/blog/uncensored-llm-leaderboard)'s rule: today's real-model check first, then output price, read live.

## The ranked list: 10 uncensored GGUF models you can also call by API

Every row is a build that exists twice: as a GGUF repo on Hugging Face, read October 27, 2026, and as an id on this key. Rows that passed today's [real-model check](https://uncensoredllm.ai/proof) rank first, then by output price per million tokens, the same rule as the [uncensored LLM leaderboard](https://uncensoredllm.ai/blog/uncensored-llm-leaderboard). 10 of 10 passed the check run at 2026-11-09 14:19 UTC. Prices are from the catalog of 2026-11-09.

| # | Model | Build | GGUF repo | Q4_K_M | Output $/M | 100 messages | Check |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | [Magidonia 24B v4.3](https://uncensoredllm.ai/models/magidonia-24b-v4.3) | Roleplay | [`TheDrummer/Magidonia-24B-v4.3-GGUF`](https://huggingface.co/TheDrummer/Magidonia-24B-v4.3-GGUF) | 14.3 GB | $0.4235 | $0.15 | Steady since Oct 28 |
| 2 | [Cydonia 24B v4.3](https://uncensoredllm.ai/models/cydonia-24b-v4.3) | Roleplay | [`TheDrummer/Cydonia-24B-v4.3-GGUF`](https://huggingface.co/TheDrummer/Cydonia-24B-v4.3-GGUF) | 14.3 GB | $0.60 | $0.21 | Steady since Oct 23 |
| 3 | [Stheno 8B v3.2](https://uncensoredllm.ai/models/stheno-8b-v3.2) | Roleplay | [`bartowski/L3-8B-Stheno-v3.2-GGUF`](https://huggingface.co/bartowski/L3-8B-Stheno-v3.2-GGUF) | 4.92 GB | $0.7035 | $0.30 | Steady since Oct 23 |
| 4 | [Anubis 70B v1.1](https://uncensoredllm.ai/models/anubis-70b-v1.1) | Roleplay | [`TheDrummer/Anubis-70B-v1.1-GGUF`](https://huggingface.co/TheDrummer/Anubis-70B-v1.1-GGUF) | 42.5 GB | $1.085 | $0.47 | Steady since Oct 28 |
| 5 | [Mag Mell R1 12B](https://uncensoredllm.ai/models/mag-mell-12b-r1) | Roleplay | [`inflatebot/MN-12B-Mag-Mell-R1-GGUF`](https://huggingface.co/inflatebot/MN-12B-Mag-Mell-R1-GGUF) | 7.48 GB | $1.7255 | $0.74 | Steady since Oct 23 |
| 6 | [NeuralDaredevil 8B Abliterated](https://uncensoredllm.ai/models/neuraldaredevil-8b-abliterated) | Abliterated | [`QuantFactory/NeuralDaredevil-8B-abliterated-GGUF`](https://huggingface.co/QuantFactory/NeuralDaredevil-8B-abliterated-GGUF) | 4.92 GB | $1.76 | $0.76 | Steady since Oct 24 |
| 7 | [UnslopNemo 12B v4.1](https://uncensoredllm.ai/models/unslopnemo-12b-v4.1) | Roleplay | [`TheDrummer/UnslopNemo-12B-v4.1-GGUF`](https://huggingface.co/TheDrummer/UnslopNemo-12B-v4.1-GGUF) | 7.48 GB | $1.972 | $0.85 | Steady since Oct 23 |
| 8 | [Llama 3.3 70B Abliterated](https://uncensoredllm.ai/models/llama-3.3-70b-abliterated) | Abliterated | [`bartowski/Llama-3.3-70B-Instruct-abliterated-GGUF`](https://huggingface.co/bartowski/Llama-3.3-70B-Instruct-abliterated-GGUF) | 42.52 GB | $2.80 | $1.20 | Steady since Oct 29 |
| 9 | [Skyfall 36B v2](https://uncensoredllm.ai/models/skyfall-36b-v2) | Roleplay | [`TheDrummer/Skyfall-36B-v2-GGUF`](https://huggingface.co/TheDrummer/Skyfall-36B-v2-GGUF) | 22.4 GB | $2.80 | $0.85 | Steady since Oct 23 |
| 10 | [Qwen3.8 27B Obliterated](https://uncensoredllm.ai/models/qwen3.8-27b-obliterated) | Abliterated | [`mradermacher/Qwen3.8-27B-OBLITERATED-GGUF`](https://huggingface.co/mradermacher/Qwen3.8-27B-OBLITERATED-GGUF) | 16.9 GB | $7.27 | $0.45 | Steady since Oct 22 |

Q4_K_M sizes as each repo page showed them on October 27, 2026. 100 messages = 100 × 4,000 input tokens at the input price plus 100 × 300 output tokens at the output price, computed at render.

[Magidonia 24B v4.3](https://uncensoredllm.ai/models/magidonia-24b-v4.3) leads today at $0.4235 per million output tokens, $0.15 per 100 messages. The smallest download on the list is Stheno 8B v3.2 at 4.92 GB for Q4_K_M.

## What each build is

From each build's Hugging Face card, as the model pages here state it, plus the note from the GGUF repo.

| Model | Base model | What the tune does | From the GGUF repo |
| --- | --- | --- | --- |
| [Magidonia 24B v4.3](https://uncensoredllm.ai/models/magidonia-24b-v4.3) | Magistral Small 2509 | Fine-tune for roleplay and story writing, by TheDrummer | Same template as Cydonia; the Magistral-based sibling, different prose. |
| [Cydonia 24B v4.3](https://uncensoredllm.ai/models/cydonia-24b-v4.3) | Mistral Small 3.2 24B Instruct 2506 | Fine-tune for roleplay and story writing, by TheDrummer | Mistral v7 Tekken template; the card points to bartowski’s iMatrix quants as the better files. |
| [Stheno 8B v3.2](https://uncensoredllm.ai/models/stheno-8b-v3.2) | Llama 3 8B | Fine-tune, sixth iteration, merged back to its base, by Sao10K | Llama 3 Instruct template; the card recommends temperature 1.12 to 1.22, min_p 0.075. |
| [Anubis 70B v1.1](https://uncensoredllm.ai/models/anubis-70b-v1.1) | Llama 3.3 70B Instruct | Fine-tune for character adherence and unalignment, by TheDrummer | Llama 3 template; two 24 GB cards or a 64 GB Mac at Q4. |
| [Mag Mell R1 12B](https://uncensoredllm.ai/models/mag-mell-12b-r1) | Mistral Nemo Base 2407 (ChatML) | Multi-stage merge: SLERP specialists, then DARE-TIES, by inflatebot | ChatML; the card was tested at temperature 1.25, min_p 0.2. |
| [NeuralDaredevil 8B Abliterated](https://uncensoredllm.ai/models/neuraldaredevil-8b-abliterated) | Daredevil 8B Abliterated (Llama 3 8B) | Abliteration, then DPO, by Maxime Labonne (mlabonne) | Abliterated, then DPO to recover the lost quality; runs on a laptop. |
| [UnslopNemo 12B v4.1](https://uncensoredllm.ai/models/unslopnemo-12b-v4.1) | Mistral Nemo 12B | Fine-tune for adventure writing and roleplay, by TheDrummer | A 12B that fits an 8 GB card at Q4. |
| [Llama 3.3 70B Abliterated](https://uncensoredllm.ai/models/llama-3.3-70b-abliterated) | Llama 3.3 70B Instruct | Abliteration, no retraining, by huihui-ai | Quants of huihui-ai’s abliteration; IQ2_M at 24.12 GB fits one 24 GB card. |
| [Skyfall 36B v2](https://uncensoredllm.ai/models/skyfall-36b-v2) | Mistral Small 2501 (upscaled) | Upscale with continued training for creativity and RP, by TheDrummer | Mistral v7 Tekken recommended; Alpaca for choose-your-own-adventure. |
| [Qwen3.8 27B Obliterated](https://uncensoredllm.ai/models/qwen3.8-27b-obliterated) | Qwen3.8 27B | Blended abliteration (SVD and LEACE), by OBLITERATUS | The author’s own repo, OBLITERATUS/Qwen3.8-27B-OBLITERATED, also carries GGUF files from Q2_K to Q8_0. |

Card facts read on October 20, 2026; the GGUF notes on October 27, 2026.

## Best uncensored GGUF for roleplay

Seven of the ten are roleplay tunes, and the GGUF lists for roleplay come down to what fits the card. At Q4_K_M: the 8B builds (Stheno, NeuralDaredevil) at about 5 GB run on a laptop; the 12B builds (UnslopNemo, Mag Mell) at about 7.5 GB fit an 8 GB card; the 24B Cydonia and Magidonia at 14.3 GB want 16 GB; Skyfall 36B at 22.4 GB wants 24 GB; the 70B rows want two cards or a 64 GB Mac. The sizes are the repos' own figures, read October 27, 2026.

By API the bill is per token, not per GB of VRAM, so the same builds appear in the table above with a price per 100 messages rather than a GB count. The [roleplay page](https://uncensoredllm.ai/roleplay) lists every tune built for cards; SillyTavern takes either route, a local server or this base URL, and the [docs](https://uncensoredllm.ai/docs#sillytavern) have the fields.

## Best abliterated LLM on the list

Three rows are abliterated rather than fine-tuned: the refusal direction was removed from a released model and nothing was added. `qwen3.8-27b-obliterated` is OBLITERATUS's edit of Qwen3.8 27B; `llama-3.3-70b-abliterated` is huihui-ai's edit of Meta's 70B; `neuraldaredevil-8b-abliterated` is mlabonne's 8B, abliterated and then given a DPO pass to recover what the edit cost. Pick one when you want a known base model's answers without the refusals. [Abliterated vs uncensored](https://uncensoredllm.ai/blog/abliterated-vs-uncensored) measures the two methods on this catalog.

## GGUF download or API: the same weights, two bills

GGUF is the single-file format llama.cpp reads, and Ollama, LM Studio, KoboldCpp and Jan load it; Hugging Face's [GGUF documentation](https://huggingface.co/docs/hub/gguf) lists the quant types, Q4_K at about 4.5 bits per weight among them. A download costs the hardware and the time to fetch, update and swap models; prompts never leave the machine. The API bills the per-token rate above.

```
# the same build two ways
ollama run hf.co/TheDrummer/Magidonia-24B-v4.3-GGUF:Q4_K_M

curl https://api.uncensoredllm.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-ullm-live-…" \
  -H "Content-Type: application/json" \
  -d '{"model": "magidonia-24b-v4.3", "messages": [{"role": "user", "content": "Hello"}]}'
```

Pick local when prompts must stay on your own hardware or a GPU already sits idle. Pick the API when there is no GPU, when several builds are worth trying by changing one id, or when the same model has to answer a phone, a server and SillyTavern at once. Both sides of this comparison in more depth are on the [leaderboard](https://uncensoredllm.ai/blog/uncensored-llm-leaderboard), which ranks every build on the key.

Live prices for every id on this list, sortable, from the catalog of 2026-11-09:

| Model | Build | Input $/M | Output $/M |
| --- | --- | --- | --- |
| [Magidonia 24B v4.3](https://uncensoredllm.ai/models/magidonia-24b-v4.3) magidonia-24b-v4.3 | Roleplay | $0.3535 | $0.4235 |
| [Cydonia 24B v4.3](https://uncensoredllm.ai/models/cydonia-24b-v4.3) cydonia-24b-v4.3 | Roleplay | $0.484 | $0.60 |
| [Stheno 8B v3.2](https://uncensoredllm.ai/models/stheno-8b-v3.2) stheno-8b-v3.2 | Roleplay | $0.7035 | $0.7035 |
| [Anubis 70B v1.1](https://uncensoredllm.ai/models/anubis-70b-v1.1) anubis-70b-v1.1 | Roleplay | $1.085 | $1.085 |
| [Mag Mell R1 12B](https://uncensoredllm.ai/models/mag-mell-12b-r1) mag-mell-12b-r1 | Roleplay | $1.7255 | $1.7255 |
| [NeuralDaredevil 8B Abliterated](https://uncensoredllm.ai/models/neuraldaredevil-8b-abliterated) neuraldaredevil-8b-abliterated | Abliterated | $1.76 | $1.76 |
| [UnslopNemo 12B v4.1](https://uncensoredllm.ai/models/unslopnemo-12b-v4.1) unslopnemo-12b-v4.1 | Roleplay | $1.972 | $1.972 |
| [Skyfall 36B v2](https://uncensoredllm.ai/models/skyfall-36b-v2) skyfall-36b-v2 | Roleplay | $1.925 | $2.80 |
| [Llama 3.3 70B Abliterated](https://uncensoredllm.ai/models/llama-3.3-70b-abliterated) llama-3.3-70b-abliterated | Abliterated | $2.80 | $2.80 |
| [Qwen3.8 27B Obliterated](https://uncensoredllm.ai/models/qwen3.8-27b-obliterated) qwen3.8-27b-obliterated | Abliterated | $0.582 | $7.27 |

Prices per million tokens, read from the live catalog on 2026-11-09.

Credit is prepaid from $10 and settles per token. The [family pages](https://uncensoredllm.ai/uncensored) link each build's Hugging Face card, so the GGUF is one click from the API page.

## How this list is ranked, and what it leaves out

Two inputs, both public. The daily check sends one fixed prompt to every model on the catalog and records whether the input token count the host reports matches that build's earlier count; a build that passed ranks above one that did not or was not checked. Among those, the lower output price per million tokens ranks higher. There is no quality score: a benchmark depends on the prompts and the judge, while the price is what the key is billed. The UGI Leaderboard scores how uncensored a model is, and the [leaderboard post](https://uncensoredllm.ai/blog/uncensored-llm-leaderboard) sets UGI's figures beside the builds it has tested.

Left out: builds on the key with no public GGUF repo found on October 27, 2026 (the GLM-5.3, DeepSeek V4 and Kimi K3 edits among them, which are run by API), and GGUF repos whose weights are not on the key (TheBloke's Luna AI Llama2 Uncensored from the 2023 lists, for one). Every repo linked above answered a page on October 27, 2026; a repo that goes private or is removed is dropped at the next review.

## Questions

### What is the best uncensored LLM GGUF?

By this list's rule, Magidonia 24B v4.3: it passed today's check and has the lowest output price of the builds that did, $0.4235 per million tokens. Its GGUF is TheDrummer/Magidonia-24B-v4.3-GGUF (Q4_K_M 14.3 GB, read October 27, 2026). For a card chat on an 8 GB GPU, the 12B rows; for a laptop, the 8B rows. See the full leaderboard.

### What are the best GGUF models for roleplay?

Cydonia 24B v4.3 and Magidonia 24B v4.3 at 16 GB of VRAM, UnslopNemo 12B and Mag Mell 12B at 8 GB, Stheno 8B on a laptop, Skyfall 36B and Anubis 70B when the card allows. All seven are roleplay tunes with the refusal trained out, and all seven answer by API on this key. See the roleplay builds.

### Is there an uncensored GGUF LLM model I can call without a GPU?

Every row of the list. Send the id with the base URL https://api.uncensoredllm.ai/v1 and a key, from a phone, a server or SillyTavern; the GGUF is the same weights for a local run when you want one. See the client setup docs.

### What is the best abliterated LLM?

Of the three abliterated rows, qwen3.8-27b-obliterated has the most downloads on its card and the widest context; llama-3.3-70b-abliterated is the strongest base; neuraldaredevil-8b-abliterated is the one that runs anywhere. The leaderboard post ranks every abliterated build on the key by today’s check and price. See abliterated vs uncensored.

### Is there an uncensored LLM leaderboard?

Two. The UGI Leaderboard on Hugging Face scores how uncensored a model is; the leaderboard here ranks the builds you can call by API by the daily check and output price, with UGI’s score beside the models it has tested. See the uncensored LLM leaderboard.

### Where do I download an uncensored GGUF?

From the repo in the table: open it on Hugging Face and pick a quant, Q4_K_M for most cards. Ollama and LM Studio pull it by name, for example ollama run hf.co/TheDrummer/Cydonia-24B-v4.3-GGUF:Q4_K_M.

### What does Q4_K_M mean?

A 4-bit quantization in llama.cpp’s K-quant family, about 4.5 bits per weight per Hugging Face’s GGUF documentation. It is the usual trade between size and quality; Q8_0 keeps more quality at roughly twice the file size, Q2_K halves it again at a visible cost.

Every model above is on the [uncensored model list](https://uncensoredllm.ai/models) with its live price per million tokens. [Per-token pricing](https://uncensoredllm.ai/pricing) sets the whole catalog side by side, and the [family directory](https://uncensoredllm.ai/uncensored) groups the builds by base model.

[Get a key](https://app.uncensoredllm.ai/console) Prepaid from $10, no subscription.

## New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.