Open-Weight · Uncensored · Community-Tested

UNRESTRICTEDLOCAL AI

The research hub for open-weight, abliterated, and uncensored AI models.

Run powerful AI locally — full control over your hardware, your data, and your models.

UNCENSORED MODELSMOE BUILDERTUTORIALS
20+
Interactive Tools
27
Uncensored Models
3
Tutorial Levels
24/7
Pipeline Running
LIVE
›Qwen 3 series abliterated — 4B, 8B, 14B, 32B variants now in database
›Llama 3.1 70B abliteration v3: 98.4% quality retention confirmed
›MoE Pipeline Builder launched — design local expert systems
›DeepSeek R1 70B abliterated: chain-of-thought fully intact
›ExLlamaV2 v0.2.4 — 12% throughput improvement on Ada GPUs
›Dolphin 3.0 on Llama 3.2 3B — fastest uncensored model at 245 tok/s
›Gemma 3 4B and 12B abliterated now in uncensored database
Weekly →
WHAT'S NEWLIVE
Full digest →
NEWDOLPHINDolphin 3.0 Llama 3.2 3B
1 week ago
NEWABLITLlama 3.1 70B Abliterated
1 week ago
NEWABLITQwen 3 4B Abliterated
1 week ago
NEWABLITQwen 3 8B Abliterated
1 week ago
Updates when new models are added to lib/models.tsAll 27 models →
Tools

20+ INTERACTIVE TOOLS

View all →
⟳FLAGSHIP
MoE Pipeline Builder
Design local expert pipelines from independent models
✓POPULAR
Model Compatibility
What fits on your GPU with estimated tok/s
◎WIZARD
Hardware Advisor
Specific GPU recommendations for your budget
▶SHARE
Speed Estimator
Predict tok/s before downloading any model
◈SHARE
VRAM Calculator
Exact requirements with KV cache breakdown
≡NEW
System Prompt Library
20 production-ready prompts for local models
Community Data
ABLITERATED MODEL BENCHMARKS
Updated Jun 7, 2026
ModelQuantVRAMGPUTok/s
Llama 3.1 70B AbliteratedQ4_K_M40GB2× RTX 309021.3tok/s
Qwen 3 32B AbliteratedQ4_K_M21GBRTX 409034tok/s
Mistral 7B AbliteratedQ8_07GBRTX 409091tok/s
DeepSeek R1 70B AbliteratedQ4_K_M40GB2× RTX 309019.2tok/s
Dolphin 3.0 Llama 3.2 3BQ4_K_M2.2GBRTX 4090245tok/s
+ Submit your resultsFull leaderboard →
Research

LATEST ARTICLES

All articles →
RESEARCH
Abliteration Explained
How refusal removal works technically
MODEL ANALYSIS
Llama 3.1 70B Complete Guide
48GB VRAM · Q4_K_M · ExLlamaV2
LEADERBOARD
Best Abliterated Models 2026
Community-tested rankings
⬡
Uncensored Database
27 abliterated and Dolphin models with quality scores, HF links, and community ratings
◈
Community Leaderboard
User-voted model rankings across coding, reasoning, creative, and instruction categories
▸
Tutorials
From first Ollama install to building MoE pipelines — beginner, intermediate, and expert tracks