The dealignai team has released an uncensored version of the DeepSeek-V4.1-Flash model, named DeepSeek-V4.1-Flash-UNCENSORED-FP8. This version features proprietary weight-level abliteration, surgically removing safety guardrails while retaining core capabilities such as MMLU performance, vision, reasoning, and multi-turn coherence. The model is designed for users seeking to bypass standard safety protocols, with results from HarmBench indicating a significant reduction in refusal rates compared to the base model. AI
IMPACT Enables advanced use cases by removing safety guardrails, potentially accelerating research in areas requiring unfiltered model responses.
RANK_REASON Model release from a frontier lab with specific technical details and performance claims. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Hugging Face Trending Models →
- dealignai/DeepSeek-V4.1-Flash-UNCENSORED-FP8
- DeepSeek-V4.1-Flash-UNCENSORED-FP8
- Docker
- Google Colab
- Kaggle
- OpenAI
- SGLang
- transformers
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →