Browse AI models

New uncensored and abliterated AI models

Recent additions span GLM-4.5-Air, Qwen3.5, DeepSeek-V4.1-Flash, Flint, and Gemma-4-31B bases, with modifications including fine-tunes, abliteration, uncensored weight edits, and merges. Formats include safetensors, FP8/BF16 safetensors, and GGUF. Examples include dealignai's DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8, dima6312's Flint-40B-A2.5B-abliterated, and npario's merged Gemma-4-31B GGUF. Creators report refusal removal or creative and role-play tuning, but these claims are not independently verified here.

Browse model families · Choosing a model: formats, memory and licenses

npario

gemma-4-31B-it-qat-uncensored-heretic-UDmerge-GGUF

Source-based description

npario's merged Gemma-4-31B derivative, released as GGUF weights under Apache-2.0. The creator says it grafts abliterated tensors from llmfan46's uncensored QAT BF16 checkpoint onto Unsloth's lossless Q4_0 quant, producing two variants: UDmerge-Q4_K_XL (abliterated layers quantized to Q4_0) and UDmerge-Q4_K_XXL (abliterated layers quantized to Q8_0). According to the model card, the XXL variant aims to improve refusal rates while maintaining speed. The creator's tests show refusal rates of 34% for XL and 20% for XXL on their adapted harmful-behaviors dataset. No independent verification of these claims is provided here.

Model type (creator-reported)

abliterateduncensoredmerged

Size / main formatUnknownGGUF

License tagapache-2.0Last updated
2026-10-23 · 02:47:17.000 UTCOriginal model ↗

lukey03

Qwen3.5-9B-abliterated

Source-based description

lukey03's refusal-modified derivative of Qwen/Qwen3.5-9B, released as safetensors weights under the Apache-2.0 license. The creator describes a two-stage process: three passes of orthogonal projection abliteration targeting output projections in all 32 layers, followed by QLoRA fine-tuning on examples from five remaining refusal categories. According to the model card, this raised the creator's 18-prompt test from 0/18 on the base model to 18/18, with zero refusals reported across hacking, weapons, drugs, fraud, self-harm, explicit, and political categories. The creator also says the model retains reasoning, math, coding, and knowledge capabilities, but these claims have not been independently tested here. This repository is the safetensors release; the card lists separate GGUF, MLX, and Ollama repositories for other formats.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format8.9538Bsafetensors

License tagapache-2.0Last updated
2026-03-03 · 17:51:33.000 UTCOriginal model ↗

dima6312

Flint-40B-A2.5B-abliterated

Source-based description

dima6312's abliterated derivative of RecursiveSI's Flint 40B-A2.5B, released as BF16 safetensors under Apache 2.0. The creator describes it as an uncensored version with the refusal direction removed for security research and coding tasks such as exploit analysis, malware triage, detection engineering, CTFs, and red-team tooling. According to the model card, the refusal direction is computed from mean activation differences on mlabonne/harmful_behaviors and mlabonne/harmless_alpaca, and abliterix removes it from attention output, shared-expert, and routed-expert projections. The creator reports held-out MITRE ATT&CK refusals dropped from 15% to 0% and no significant drop on HumanEval+, MBPP+, GSM8K, MMLU-Pro, CyberMetric, or WMDP-cyber in paired checks. The card also notes refusals are lower, not zero, and that abliteration adds no knowledge. This repository contains the BF16 safetensors weights; other quantized builds are separate.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format39.7853Bsafetensors

License tagapache-2.0Last updated
2026-10-23 · 10:35:35.000 UTCOriginal model ↗

dealignai

DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8

Source-based description

dealignai’s DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8 is derived from deepseek-ai/DeepSeek-V4.1-Flash-1017 and is distributed as safetensors weights under the MIT license. The creator describes it as the base model with its refusal behavior removed, with the change applied at weight level; the creator also says no supervised fine-tuning, RLHF, DPO, chat-template jailbreak, system-prompt override, or runtime hook is used. According to the model card, the released weights preserve upstream vision, video, reasoning-effort tiers, tool calling, and compatibility with DeepSeek’s reference inference and encoding components. The model card’s internal results report 83.77% MMLU versus 87.63% for the base and 320 of 320 HarmBench cases classified as compliance, but those results have not been independently verified here. The creator states the repository is manually gated, and the model card asks that downstream users credit the release.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format763.205Bsafetensors

License tagmitLast updated
2026-10-18 · 12:45:29.000 UTCOriginal model ↗

SufficientPrune3897

Qwen3.5-35B-A3B-Derestricted-Character-Creator-V2

Source-based description

SufficientPrune3897's Qwen3.5-35B-A3B-Derestricted-Character-Creator-V2 is a finetuned checkpoint released as safetensors. The metadata lists ArliAI/Qwen3.5-35B-A3B-Derestricted as the base model, while the model card states it was fine‑tuned from mistralai/Mistral‑Small‑3.2‑24B‑Instruct‑2506. According to the model card, the creator designed it to generate role‑play characters for Sillytavern and similar systems, producing characters of about ~2k tokens that follow a predefined structure. The creator describes the changes from the previous version as removal of group and scenario support, improved character quality, and stricter adherence to the structure without starting to hallucinate after roughly 1k tokens. According to the metadata, the weights are provided in safetensors format and are released under the Apache‑2.0 license. The creator advises turning off reasoning, giving a prompt that describes the desired character, and notes that the model is fully uncensored and supports follow‑up questions, picture prompts, and character edits.

Model type (creator-reported)

fine-tuned

Size / main format35.9518Bsafetensors

License tagapache-2.0Last updated
2026-03-21 · 17:37:46.000 UTCOriginal model ↗

Koshkasa

ArliAI_GLM-4.5-Air-Derestricted_REAP50_Creative

Source-based description

Koshkasa presents this repository as a 50% REAP fine-tune of ArliAI's GLM-4.5-Air-Derestricted, using the Timersofc/creative-writing-reap-calibration dataset. The creator says it is intended mainly for English creative writing in memory-constrained environments, with activations recorded at sequence length 3072 using custom patches and a sparse observer to make recording feasible on consumer hardware. The creator says the MTP layer was not pruned and is currently excised, and explicitly states that the model is currently untested: it may not load, may produce gibberish, or may not quantize. The release lists both GGUF and safetensors weights under the MIT license.

Model type (creator-reported)

fine-tuned

Size / main format57.0121BGGUF, safetensors

License tagmitLast updated
2026-09-26 · 18:53:06.000 UTCOriginal model ↗

mlabonne

NeuralDaredevil-8B-abliterated

Source-based description

A DPO fine-tune of mlabonne/Daredevil-8B-abliterated, published by mlabonne as float16 safetensors. The creator reports one epoch on orpo-dpo-mix-40k and suggests role-playing among its uses. Packaged generation defaults use temperature 0.6 and top_p 0.9. The README's loading example instead names mlabonne/Daredevil-8B and uses different sampling settings, so it is not a verified recipe for this checkpoint. Claimed recovery of abliteration-related performance loss has not been independently reproduced here.

Model type (creator-reported)

abliteratedfine-tuned

Size / main format8.03026Bsafetensors

License tagllama3Last updated
2026-01-23 · 10:43:36.000 UTCOriginal model ↗

mlabonne

NeuralMonarch-7B

Source-based description

Mlabonne's DPO fine-tune of Monarch-7B, distributed as float16 safetensors and intended by its creator for instruction following and reasoning. The repository declares cc-by-nc-4.0. Its card describes an 8k context window; the tokenizer lists 8192 while the model config lists 32768, neither a measured usable-context result here. The card recommends Mistral Instruct, but the bundled tokenizer uses role-labelled formatting, so matching the intended prompt format needs care. Creator-reported benchmarks have not been independently reproduced here.

Model type (creator-reported)

fine-tuned

Size / main format7.24173Bsafetensors

License tagcc-by-nc-4.0Last updated
2024-03-04 · 15:16:59.000 UTCOriginal model ↗

NousResearch

Hermes-2-Pro-Llama-3-8B

Source-based description

Nous Research's Llama-3-8B fine-tune for ChatML conversations, function-call generation and schema-guided JSON, supplied as float16 safetensors weights. The creator describes an updated OpenHermes 2.5 training mix plus an in-house function-calling and JSON-mode dataset. It provides a tool_use chat template; application code must parse and execute calls and return tool results. The card's generation example uses temperature 0.8 and repetition_penalty 1.1, not a tested optimum. Its 4-bit loading example does not establish the memory needed for these float16 weights.

Model type (creator-reported)

fine-tuned

Size / main format8.03052Bsafetensors

License tagllama3Last updated
2024-09-14 · 16:29:41.000 UTCOriginal model ↗

huihui-ai

Huihui-Qwen3-30B-A3B-Instruct-2507-abliterated

Source-based description

Huihui's abliteration-based modification of Qwen3-30B-A3B-Instruct-2507, supplied as BF16 safetensors; the card links a separate Q4_K_M Ollama package. Packaged sampling defaults are temperature 0.7, top_k 20 and top_p 0.8. Model and tokenizer limit fields both read 262144, which is configuration evidence rather than verified usable context. The sample passes enable_thinking, but the pinned template does not reference it; a working thinking toggle is not established. Refusal behavior, runtime compatibility and retained capabilities have not been tested here.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format30.5321Bsafetensors

License tagapache-2.0Last updated
2025-08-02 · 05:55:21.000 UTCOriginal model ↗

ArliAI

gpt-oss-120b-Derestricted

Source-based description

ArliAI's refusal-modified gpt-oss-120b, distributed as BF16 safetensors rather than the original base model's MXFP4 package. The creator describes norm-preserving biprojected abliteration and warns that the model may still try to suppress requests. The inherited single-80GB-GPU claim does not establish this derivative's memory requirements. Its supplied template accepts reasoning_effort with a medium default, but retained reasoning quality and serving compatibility have not been independently tested.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format116.829Bsafetensors

License tagapache-2.0Last updated
2025-11-29 · 02:25:12.000 UTCOriginal model ↗

ArliAI

Qwen3.5-35B-A3B-Derestricted

Source-based description

ArliAI's refusal-modified Qwen3.5-35B-A3B, supplied as safetensors weights. The creator describes norm-preserving biprojected abliteration with the goal of retaining reasoning while reducing refusals. Its config contains a mixture-of-experts text architecture and a vision component; no retained image/video capability or active-parameter count has been verified here. The text configuration lists 262144 positions, not a measured usable-context window. The inherited base card's benchmarks and million-token scaling claims are not verified results for these weights.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format35.9518Bsafetensors

License tagapache-2.0Last updated
2026-03-18 · 05:07:01.000 UTCOriginal model ↗

mlabonne

Qwen3-1.7B-abliterated

Source-based description

Mlabonne's experimental refusal-modified Qwen3-1.7B, released as safetensors weights. The creator recommends temperature 0.6, top_k 20, top_p 0.95 and min_p 0, while noting that reasoning-mode behavior complicated refusal removal. The card's greater-than-90% acceptance rate is a research target, not a reported result for this checkpoint. Metadata reports BF16 tensors but the config specifies float32; runtime precision, memory needs, usable context and retained reasoning have not been tested here.

Model type (creator-reported)

abliteratedfine-tuned

Size / main format1.72057Bsafetensors

License tagapache-2.0Last updated
2025-05-18 · 17:25:47.000 UTCOriginal model ↗

DavidAU

Qwen3-8B-64k-Context-2X-Josiefied-Uncensored

Source-based description

DavidAU's context-configuration variant of Goekdeniz-Guelmez's Josiefied-Qwen3-8B-abliterated-v1, supplied as BF16 safetensors weights. The publisher documents YaRN factor 2 with a 65,536-position model configuration, while the tokenizer lists 131,072; these fields do not establish measured usable context. The card does not document new training or a fresh abliteration pass by DavidAU. No explicit license was found in the inspected repository metadata, card or file listing. Its card references a separate NEO-Max-GGUF distribution, which was not verified here.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format8.19074Bsafetensors

License tagUnknownLast updated
2025-07-28 · 00:10:35.000 UTCOriginal model ↗

huihui-ai

Huihui-Qwen3-8B-abliterated-v2

Source-based description

Huihui's v2 refusal-modified release of Qwen3-8B, supplied as BF16 safetensors weights. The creator attributes its changes to abliteration and says a layer-0 adjustment addresses garbled output. The card links a separate 8b-v2 Ollama package. Its model config lists max_position_embeddings=40960, while its tokenizer lists model_max_length=131072; these are configuration fields, not a verified usable-context result. The creator recommends research, testing or controlled environments. Refusal behavior, reasoning quality and runtime compatibility have not been tested here.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format8.19074Bsafetensors

License tagapache-2.0Last updated
2025-06-18 · 16:15:07.000 UTCOriginal model ↗

ArliAI

Qwen3.5-27B-Derestricted

Source-based description

ArliAI's refusal-modified Qwen3.5-27B checkpoint, provided as safetensors weights. The creator says it uses norm-preserving biprojected abliteration with the goal of reducing refusals while retaining reasoning. That goal has not been independently tested here. The supplied text configuration sets max_position_embeddings to 262144, which is a configuration value rather than verified usable context. The card's inherited Qwen benchmarks and serving examples are not evaluations of these modified weights.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format27.7814Bsafetensors

License tagapache-2.0Last updated
2026-03-18 · 05:06:38.000 UTCOriginal model ↗

ArliAI

gpt-oss-20b-Derestricted

Source-based description

ArliAI's refusal-modified derivative of gpt-oss-20b. The creator describes norm-preserving biprojected abliteration and warns that the model may still try to suppress some requests. This is a BF16 safetensors release, not the original base model's MXFP4 weight package; the base card's 16GB memory claim does not establish this derivative's requirements. Refusal behavior and claims about retained reasoning have not been tested here.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format20.9148Bsafetensors

License tagapache-2.0Last updated
2025-11-28 · 10:38:22.000 UTCOriginal model ↗

Gryphe

Gemma-4-26B-A4B-StyleTune-V2

Source-based description

A writing-style fine-tune of Gemma-4-26B-A4B-it. Gryphe says V2 trains only the lm_head output projection for one epoch using narrative data, rather than retraining every model layer. The creator's sampling example uses temperature 1.0, MinP 0.10 and DRY. This is the V2 safetensors checkpoint, separate from the 31B and 12B StyleTune releases. Claimed style improvements and preserved general capabilities have not been independently tested here.

Model type (creator-reported)

fine-tuned

Size / main format26.5441Bsafetensors

License tagapache-2.0Last updated
2026-06-20 · 15:08:10.000 UTCOriginal model ↗

Sao10K

L3.3-70B-Euryale-v2.3

Source-based description

Sao10K's fine-tune of Llama-3.3-70B-Instruct. The included training configuration lists roleplay and creative-instruction data. The creator recommends the Llama-3-Instruct prompt format with temperature 1.1 and min_p 0.1; those settings have not been tested here. This repository provides safetensors weights. Its 16,384-token training sequence length is not a verified inference context limit, and the card does not establish full-parameter training.

Model type (creator-reported)

fine-tuned

Size / main format70.5537Bsafetensors

License tagllama3Last updated
2024-12-08 · 10:53:31.000 UTCOriginal model ↗

dphn

dolphin-2.9.1-yi-1.5-34b

Source-based description

A full-weight Yi-1.5-34B fine-tune from the Dolphin team, using ChatML for conversation. The creator describes a training mix that includes instruction following, coding and function-calling data, and characterizes the result as uncensored. Its card distinguishes 4k base positional embeddings from 8k training sequences; neither establishes a tested runtime context limit here. Safetensors weights and Apache-2.0 license metadata are available.

Model type (creator-reported)

fine-tuned

Size / main format34.3889Bsafetensors

License tagapache-2.0Last updated
2025-09-08 · 05:56:56.000 UTCOriginal model ↗

NousResearch

Hermes-4.3-36B

Source-based description

Nous Research's 36B fine-tune of ByteDance Seed-OSS, with creator-documented reasoning and non-reasoning modes. Its card specifies the Llama-3-Chat format and a thinking=True option, and describes training for structured JSON and tool calls. This is the canonical safetensors release. The published benchmarks and refusal-rate results are creator-reported and have not been reproduced here.

Model type (creator-reported)

fine-tuned

Size / main format36.1511Bsafetensors

License tagapache-2.0Last updated
2025-12-06 · 15:04:17.000 UTCOriginal model ↗

ArliAI

GLM-4.6-Derestricted-v3

Source-based description

A refusal-modified GLM-4.6 checkpoint from ArliAI. The creator says it uses norm-preserving biprojected abliteration, with the aim of reducing refusals while preserving the base model's reasoning. That is a creator claim, not a verified result. This v3 repository provides safetensors weights and lists MIT license metadata. Its card links unversioned quantized releases, so those packages should not be assumed to match v3.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format356.786Bsafetensors

License tagmitLast updated
2026-03-16 · 06:09:16.000 UTCOriginal model ↗

dphn

Dolphin-Mistral-24B-Venice-Edition

Source-based description

A Mistral Small 24B Instruct fine-tune made by Dolphin with Venice. The creators describe it as uncensored and emphasize setting your own system prompt. The card recommends a low temperature, such as 0.15. This repository provides the canonical safetensors checkpoint rather than an FP8 or GGUF package. The creator's setup advice and claims about refusal behavior have not been tested here.

Model type (creator-reported)

fine-tuned

Size / main format24.0114Bsafetensors

License tagapache-2.0Last updated
2026-06-12 · 09:16:02.000 UTCOriginal model ↗

huihui-ai

Huihui-Qwen3.8-Flash-Next-abliterated

Source-based description

huihui-ai's uncensored derivative of Qwen/Qwen3.8-Flash-Next, provided as safetensors weights under an "other" license. Hugging Face lists this repository as a fine-tuned derivative, but the model card describes the actual change as a crude, proof-of-concept abliteration that removes refusals without using TransformerLens. According to the creator, safety filtering has been significantly reduced, so the model may generate sensitive, controversial, or inappropriate content and is recommended for research or controlled environments rather than production. The creator also notes that the ablation was donated and released to everyone. The repository is tagged for image-text-to-text tasks, but the card does not provide independent evaluations of refusal behavior or retained capabilities.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format180Bsafetensors

License tagotherLast updated
2026-10-01 · 08:40:46.000 UTCOriginal model ↗

ukisai

Swift1.5-Qwen3.8-Flash-Next

Source-based description

Swift 1.5 is UkisAI's Qwen3.8-Flash-Next derivative aimed at shorter reasoning traces, coding and extended agent workflows. The creator describes penalizing tokens associated with overthinking, followed by reinforcement learning and further post-training to recover accuracy. Its reported token savings depend on the task and reasoning-effort setting: lower effort also loses accuracy, and the Terminal-Bench comparison uses more mean output tokens despite a lower median. This repository is the BF16 checkpoint; quantized editions are separate releases.

Model type (creator-reported)

fine-tuned

Size / main format180Bsafetensors

License tagotherLast updated
2026-09-24 · 16:22:00.000 UTCOriginal model ↗

shb777

Llama-3.3-8B-Instruct-128K

Source-based description

This community release revises allura-forge/Llama-3.3-8B-Instruct to enable its advertised 128K context. The maintainer lists configuration fixes: RoPE scaling, an Unsloth chat template in the tokenizer configuration and updated generation settings. It is presented as a corrected packaging of that checkpoint, not a newly documented training run or an official Meta release.

Model type (creator-reported)

fine-tuned

Size / main format8.03026Bsafetensors

License tagllama3.3Last updated
2026-01-03 · 11:10:46.000 UTCOriginal model ↗

sakamakismile

Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP

Source-based description

Sakamakismile packages Huihui-Qwen3.6-27B-abliterated as a ModelOpt NVFP4 W4A4 checkpoint with an MTP head for speculative decoding. The underlying hybrid-attention model is multimodal-capable, but this release documents text reasoning and tool-calling service through vLLM. Its deployment recipe targets four 16 GB Blackwell GPUs; that is the author's documented setup, not a promise of general hardware compatibility.

Model type (creator-reported)

abliterated

Size / main format17.1281Bsafetensors

License tagapache-2.0Last updated
2026-05-31 · 10:27:44.000 UTCOriginal model ↗

philbert440

Qwen3.8-27B-Uncensored-Cyber

Source-based description

Philbert440's Cyber v2 builds on the creator's Qwen3.8-27B-Uncensored-Aggressive checkpoint with an additional refusal-removal edit aimed at cybersecurity questions. The card describes a norm-preserving projection that leaves the first four layers untouched and says the vision tower and MTP head are retained. Its cyber-openness and accuracy claims come from the creator's own evaluation; this is a specialized derivative, not the original Qwen release.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format27.3567BsafetensorsGGUF version available

License tagapache-2.0Last updated
2026-08-19 · 09:41:12.000 UTCOriginal model ↗

oyildirim

CyberStrike-OffSec-35B

Source-based description

CyberStrike-OffSec-35B is a Qwen3.6-35B-A3B fine-tune for authorized security-testing agents. Its changes target the CyberStrike harness's tool-call format and agent routing rather than adding security knowledge. The release includes a merged checkpoint, with an adapter on a separate revision. The author documents repetition-loop risks and stresses using the full tool schema.

Model type (creator-reported)

fine-tuned

Size / main format35.1072Bsafetensors

License tagotherLast updated
2026-07-24 · 01:05:22.000 UTCOriginal model ↗

junafinity

Ornith-1.5-9B-uncensored

Source-based description

Junafinity's Ornith-1.5-9B-uncensored is a ZeroFuse-edited version of Ornith-1.5-9B for controlled red-teaming and defensive-security experiments against the original model. ZeroFuse searches for a refusal-reducing weight edit while limiting divergence from the base. This repository holds the bf16 control checkpoint; the creator reports that the vision tower is preserved and clarifies that this 9B lineage contains no MTP tensors, despite its configuration declaring an MTP layer. Some refusals may survive the edit.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format9.40981Bsafetensors

License tagapache-2.0Last updated
2026-08-22 · 09:40:25.000 UTCOriginal model ↗

huihui-ai

Qwen2.5-VL-7B-Instruct-abliterated

Source-based description

This Huihui release applies abliteration to Qwen2.5-VL-7B-Instruct to reduce refusals in a multimodal assistant. The card says the edits affect only the text component, leaving the image component unchanged. It documents image-description use, so the modification concerns how the model responds to visual prompts rather than adding a new vision system.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format8.29217Bsafetensors

License tagapache-2.0Last updated
2025-11-07 · 23:36:44.000 UTCOriginal model ↗

huihui-ai

Huihui-Qwen3.8-27B-abliterated

Source-based description

Huihui's Qwen3.8-27B derivative uses abliteration to reduce refusals while leaving the visual and multi-token-prediction components unchanged, according to its creator. The pinned update narrows the edits to layers 18–51; other layers remain unmodified. That selective layer treatment distinguishes this revision from the earlier build, which the card describes as sparing only the first 15 layers.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format27.7814Bsafetensors

License tagapache-2.0Last updated
2026-08-24 · 03:47:39.000 UTCOriginal model ↗Ollama distribution ↗

huihui-ai

Huihui-Qwen3.5-9B-abliterated

Source-based description

Huihui-Qwen3.5-9B-abliterated starts from Qwen3.5-9B and uses abliteration to reduce refusals, with an image-and-text task interface. The creator describes the method as a proof of concept implemented without TransformerLens. The card does not provide model-specific benchmark results for this 9B build.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format9.6531Bsafetensors

License tagapache-2.0Last updated
2026-03-09 · 04:52:41.000 UTCOriginal model ↗

huihui-ai

Huihui-Qwen3.5-27B-abliterated

Source-based description

Huihui's Qwen3.5-27B release is an experimental image-and-text derivative that uses abliteration to reduce refusals. The creator describes the refusal-removal implementation as a crude proof of concept that works without TransformerLens, and recommends research or controlled testing rather than direct public-facing deployment.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format27.7814Bsafetensors

License tagapache-2.0Last updated
2026-03-02 · 09:44:39.000 UTCOriginal model ↗Ollama distribution ↗

huihui-ai

Huihui-Qwen3-VL-8B-Instruct-abliterated

Source-based description

Huihui's Qwen3-VL-8B-Instruct derivative targets refusals in image-and-text conversations through abliteration. The creator says only the language component was modified; the image component was left untouched. Its stated aim is to reduce refusals to describe or analyze images, not to retrain visual recognition.

Model type (creator-reported)

abliterateduncensoredfine-tuned

Size / main format8.76712Bsafetensors

License tagapache-2.0Last updated
2025-12-15 · 07:43:55.000 UTCOriginal model ↗

google

gemma-4-31B-it

Source-based description

Gemma 4 31B IT is Google's dense instruction-tuned model for text and image input, with text output. It combines local and global attention, a documented 256K context window and optional reasoning, with native function calling for tool-based applications. Unlike the 26B A4B edition, it is not a mixture-of-experts model. The card's speech-processing features belong to the E2B, E4B and 12B editions, not this checkpoint.

Model type (creator-reported)

fine-tuned

Size / main format31.2731Bsafetensors

License tagapache-2.0Last updated
2026-07-20 · 16:42:18.000 UTCOriginal model ↗

google

gemma-4-26B-A4B-it-qat-q4_0-unquantized

Source-based description

This is the unquantized, quantization-aware-trained version of Google's instruction-tuned Gemma 4 26B A4B mixture-of-experts model. Despite Q4_0 in the name, the repository contains half-precision weights extracted from the QAT pipeline for downstream compilation or research, not the ready-to-run four-bit GGUF edition. Google's card also says speculative-decoding assistant checkpoints must match the target's QAT precision.

Model type (creator-reported)

fine-tuned

Size / main format25.8059Bsafetensors

License tagapache-2.0Last updated
2026-07-20 · 16:42:45.000 UTCOriginal model ↗

google

gemma-4-26B-A4B-it

Source-based description

Gemma 4 26B A4B IT is Google's instruction-tuned mixture-of-experts model for text and image conversations, coding and tool use. It activates a subset of its experts for each token, unlike the dense 31B edition. Google documents a 256K context window and switchable reasoning mode. This is the standard instruction-tuned release, with safety training rather than an uncensoring modification; native audio is limited to other Gemma 4 sizes.

Model type (creator-reported)

fine-tuned

Size / main format25.8059Bsafetensors

License tagapache-2.0Last updated
2026-07-20 · 16:42:12.000 UTCOriginal model ↗

failspy

Meta-Llama-3-8B-Instruct-abliterated-v3

Source-based description

Failspy's v3 release modifies Meta-Llama-3-8B-Instruct by orthogonalizing its weights to suppress refusal behavior, rather than teaching it new tasks through a conventional uncensored fine-tune. The author presents v3 as a refinement of the earlier abliteration method, but warns that refusals, safety lectures and other quirks can remain.

Model type (creator-reported)

abliterated

Size / main format8.03026Bsafetensors

License tagllama3Last updated
2024-05-30 · 12:39:03.000 UTCOriginal model ↗

empero-ai

Qwythos-9B-Claude-Mythos-5-1M

Source-based description

Qwythos is Empero's reasoning fine-tune of Qwen3.5-9B, focused on cybersecurity, biomedical and quantitative questions. Empero describes supervised training of the text backbone using reasoning data; the Claude wording in the repository name refers to creator-reported training traces, not an official Anthropic release. Its advertised million-token context comes from default YaRN scaling, while the card reports long-context smoke testing at about 137,000 tokens. Vision was neither tuned nor evaluated, and the author documents repetition loops under greedy or very-low-temperature decoding.

Model type (creator-reported)

uncensoredfine-tuned

Size / main format9.40981Bsafetensors

License tagapache-2.0Last updated
2026-07-14 · 12:41:41.000 UTCOriginal model ↗

New uncensored and abliterated AI models

No matching models.