Community releases abliterated Qwen3.8 model with 4,600 downloads

A community abliterated build of Qwen3.8-35B-A3B-Distill-MTP-APEX-I-MiniPlus-V2.1 by IsValorum has appeared on Hugging Face, downloaded about 4,600 times. The APEX-I-MiniPlus V2.1 quantization claims a full 256K context on a 24GB GPU and lists Thai among supported languages.

• The model has been downloaded about 4,600 times on Hugging Face
• It claims a 256K context on a 24GB GPU
• A UNLV study compared four abliteration tools on 16 models from 7B to 14B
• ErisForge cut GSM8K by only 0.28 percentage points