Updated November 18, 2026
Best abliterated model to run locally
Qwen 3.8 27B Abliterated is the best abliterated model to run locally, on one 24 GB GPU.
The picks
By GPU
| Your hardware | Run | Size at 4-bit |
|---|---|---|
| 8 GB GPU or a laptop | Llama 3.1 8B Abliterated | ~5 GB |
| 24 GB GPU | Qwen 3.8 27B Abliterated | ~17 GB |
| 24 GB GPU, faster replies | Qwen 3.6 35B-A3B Abliterated | ~22 GB |
| No GPU, or you need 1M tokens | DeepSeek V4.1 Flash Abliterated by API | Hosted |
Models above about 35B parameters, such as GLM-5.3 or Kimi K3.1 Pro, need a multi-GPU server, and Silk Compute serves both by API.