GPT-OSS 120b Uncensored Aggressive Release (MXFP4 GGUF)
This aggressive uncensored model bypasses all safety filters while maintaining full capabilities.
Deep Dive
A developer has released an uncensored version of the GPT-OSS 120B model, specifically engineered to show "0 refusals to any query." The 117B parameter model uses a Mixture of Experts architecture with 128 experts and supports a 128K context window. Available as a single 61GB file, it's designed to run on hardware like an H100 GPU. The creator emphasizes "effectively lossless uncensoring" with no dataset changes or capability loss, making it highly accessible for local deployment.
Why It Matters
It represents a significant step in removing AI safety guardrails, making powerful, unrestricted models widely available for local use.