2026/11/18The November 2026 edition ranks 6 APIs and 8 models
Hosted and local, ranked by the editors

Best abliterated models

The best abliterated model is Kimi K3.1 Pro Abliterated. DeepSeek V4.1 Flash Abliterated is the best value, and Qwen 3.8 27B Abliterated is the best to run at home.

Updated November 18, 20268 modelsSizes from Hugging Face

01

Kimi K3.1 Pro Abliterated

The largest model ranked, at 2,780B parameters, with 1M tokens of context and video input.

Best overallSecurity researchLong agent runs
2,780BHosted API
from $3.00 / $8.00
02

DeepSeek V4.1 Flash 1017 Abliterated

A 763B model with 1M tokens of context at the lowest blended price ranked.

Best valueCoding agentsAll-day use
763BHosted API
from $0.24 / $0.90
03

GLM-5.3 Abliterated

Z.ai's 753B text-only model, strong at software work.

CodingLong documents
753BHosted API
from $2.00 / $4.00
04

GLM-5.3 Flash Abliterated

A 320B mixture-of-experts model with 18B active parameters, for classification and data generation at volume.

Lowest input priceHigh volume
320BHosted API
from $0.24 / $1.00
05

Qwen 3.8 27B Abliterated

The best abliterated model to run at home, at about 17 GB at 4-bit on one 24 GB GPU.

Best local model24 GB GPU
27.8B24 GB GPU
from $1.60 / $12.00
06

Gemma 4 31B Heretic

Google's Gemma 4 31B, abliterated by Heretic, the open-source tool that searches for abliteration settings.

24 GB GPUWriting
31.3B24 GB GPU
07

Qwen 3.6 35B-A3B Abliterated

A mixture-of-experts model that runs 3B of its 36B parameters per token, so each token needs little compute.

Fast on a home GPUMixture of experts
36B24 GB GPU
08

Llama 3.1 8B Abliterated

The 2024 build by mlabonne, whose write-up made abliteration popular, and small enough for a laptop.

Laptop8 GB GPU
8B8 GB GPU or laptop