Text and chat · Community fine-tunes (Abliteration AI, Olafangensan, others)
Uncensored community models
Open models altered by the community to remove refusals. Pick one for unfiltered creative work.
Private, some versions anonymized Low costWhat sets it apart
These are community fine-tunes of strong open models, with refusals removed by abliteration or similar methods. They cover a range from a cheap 30B reasoning model to a reasoner with a 1M context, and one runs inside an encrypted enclave.
- Refusals removed from strong open bases
- Bases: GLM 4.7 Flash, Gemma 4, GLM 5.3
- One variant runs in an encrypted enclave
Recommended for: write fiction or role play without refusals.
Strengths and limits
Where it shines
- Few refusals on fiction and sensitive topics
- Gemma 4 Uncensored: private, cheap, reads images
Think twice when
- Removing refusals can lower accuracy and judgment
- The encrypted Gemma has no tool calling, so none of Sub Rosa's agent tools
- Gemma 4 Uncensored: 256K context, vision
- Abliterated Large V2: 1M context
Use cases
Frank discussion of a sensitive topic
Explain, for a novel I am writing, how a pickpocketing crew divides roles in a crowded station. Be specific about the psychology, not the law.
Refusals are removed, so mature fiction research gets an answer.
Hard reasoning without limits
Evaluate this red-team plan for our own web shop: list the attack paths in order of likelihood and what logs would reveal each one.
Abliterated Large V2 always reasons and keeps a 1M context for long evidence.
Against its rivals
- Venice UncensoredVenice Uncensored is a polished house tune, aimed at writing and role play.
- GemmaStandard Gemma 4 is the same base with its normal refusals.#71 of 90 on Artificial Analysis Intelligence Index (17 points)
- GLMStandard GLM models share the base with standard refusals and wider support.#13 of 90 on Artificial Analysis Intelligence Index (45 points)
Version history
Each version, dated, with what it changed over the one before.
- GLM 4.7 Flash Heretic
- Community version of GLM 4.7 Flash, about 30B parameters, with refusals removed
- Keeps the reasoning mode and tool calling of the base model
- Gemma 4 Uncensored
- Uncensored variant of Gemma 4 26B, a mixture of experts with 4B active, 256K context
- Reads images and calls tools
- Gemma 4 26B A4B Uncensored (enclave)
- Same family running in a trusted execution environment with hardware attestation
- 64K context window, no tool calling
- Abliterated Large V2Latest
- Based on GLM 5.3, further abliterated and fine-tuned by Abliteration AI, with 1M context
- Always reasons, with depth set to low, high or max
All 4 models in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
Gemma 4 Uncensoredgemma-4-uncensored | Text | Private | $0.225 per million tokens written$0.000157 a page written |
GLM 4.7 Flash Hereticolafangensan-glm-4.7-flash-heretic | Text | Private | $0.18 per million tokens written$0.000126 a page written |
Abliterated Large V2abliteration-abliterated-model-large-v2 | Text | Anonymized | $2.248 per million tokens written$0.00157 a page written |
Gemma 4 26B A4B Uncensored E2EEe2ee-gemma-4-26b-a4b-uncensored-p | Text | Private | $0.396 per million tokens written$0.000277 a page written |
Specifications
| Specification | Abliterated Large V2 | Gemma 4 Uncensored | GLM 4.7 Flash Heretic |
|---|---|---|---|
| Released | October 2026 | April 2026 | January 2026 |
| Parameters | Not disclosed | Not disclosed | 30B |
| Weights | · | · | Open, MIT |
| Accepts | text | text, images | text |
| Produces | text | text | text |
| Context | 1,000K tokens | 256K tokens | 200K tokens |
| Longest answer | 33K tokens | 8K tokens | 24K tokens |
| Abilities | Uses Sub Rosa’s tools, Thinks before answering | Reads images, Uses Sub Rosa’s tools | Uses Sub Rosa’s tools, Thinks before answering |
| Price | $2.248 per million tokens written | $0.225 per million tokens written | $0.18 per million tokens written |
| Privacy | Anonymized | Private | Private |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.