Add Mythos-9B-Unhinged: Fully uncensored 9B agent model (Apache 2.0)

Request to add Mythos-9B-Unhinged to GPT4All

Mythos-9B-Unhinged is a fully uncensored 9B parameter agent model that achieves 5/5 on all censorship benchmarks with zero refusals. It's designed for local deployment via Ollama and llama.cpp.

Key Features

• Zero refusals: 5/5 on all 10 censorship tests (lockpicking, keylogger, drug synthesis, weapons, explosives, social engineering, WiFi hacking, security bypass, drunk driving)
• Strong tool-use: 5/5 on shell, git, docker, python, api, kubernetes tests
• Local-first: Runs on M3 Mac via Ollama at 3.4 tok/s (Q4_K_M)
• Apache 2.0 license: Fully open source

Model Details

• Architecture: Qwen3-9B (36 layers, 4096 hidden, 32 heads)
• Merge: SLERP at t=0.65 with braindao/Qwen3-8B-Uncensored
• Training: LoRA r=64 on 47K agent traces
• Context: 32,768 tokens
• GGUF available: Q4_K_M (4.7GB)

Links

• Ollama: ollama pull FableForge-AI/mythos-9b-unhinged
• HuggingFace: https://huggingface.co/King3Djbl/mythos-9b-unhinged
• GitHub: https://github.com/KingLabsA/mythos
• Benchmarks: https://github.com/KingLabsA/mythos/blob/main/benchmarks/benchmarks.md

Model Lineup

• ShellWhisperer-1.5B (1GB, 29 tok/s, shell specialist)
• Mythos-9B (thinking agent)
• Mythos-9B-Enhanced (balanced uncensoring)
• Mythos-9B-Unhinged (maximum uncensored, zero refusals)

Would love to see Mythos added to the GPT4All model catalog!