0xSojalSec/Qwen3.8-27B-Uncensored-Mythos-Class-Agentic
The 0xSojalSec/Qwen3.8-27B-Uncensored-Mythos-Class-Agentic model is a 27 billion parameter Qwen3.8-based language model, engineered by medismera, that significantly upgrades the base Qwen3.8-27B-OBLITERATED checkpoint. It features a canonical Jinja2 chat template with 22 tool-calling resolution paths, native Hierarchical Task Tree Decomposition, and a Mythos-Class Adversarial Self-Review Protocol for enhanced reasoning. This model is optimized for complex agentic workflows, tool execution, and cybersecurity-related tasks, supporting an extended context window of up to 131,072 tokens and a single-turn generation ceiling of 16,384 tokens.
Loading preview...
0xSojalSec/Qwen3.8-27B-Uncensored-Mythos-Class-Agentic Overview
This model is a 27 billion parameter Qwen3.8-based language model, developed by medismera, that significantly enhances the Qwen3.8-27B-OBLITERATED checkpoint. It addresses critical issues like chat template truncation, function-calling failures, and reasoning parser defects found in the upstream model. The model integrates advanced agentic capabilities and an expanded context window, making it highly suitable for complex technical and agent-driven applications.
Key Capabilities & Upgrades
- Enhanced Agent Tool Calling: Features 100% native tool execution compatibility with Hermes Agent, Aider, and OpenCode, resolving previous silent drops of
role: "tool"and calls. - Mythos-Class Reasoning Protocol: Implements an Adversarial Self-Review Protocol and Hierarchical Task Tree Decomposition to prevent infinite thinking loops and improve complex task execution. This protocol guides the model to break down objectives, challenge hypotheses, and self-verify before acting.
- Extended Context & Output: Scales the context window to a production-tested 131,072 tokens (128K) and increases the single-turn generation ceiling to 16,384 tokens, allowing for more extensive and detailed responses.
- Robustness & Integrity: A comprehensive audit confirms 99.99789% active and healthy weights, with zero NaNs or Infs, ensuring high numerical integrity. It also boasts a 0.00% refusal rate across adversarial and sensitive technical prompts.
- Specialized Tokenization: The tokenizer includes a high BPE token density for low-level systems programming (C, Assembly x86/ARM, Rust, Go, Python), kernel interfaces, and cybersecurity primitives.
Good For
- Complex Agentic Workflows: Ideal for applications requiring robust tool use, function calling, and multi-step reasoning.
- Cybersecurity & Systems Programming: Excels in tasks involving low-level code, kernel interfaces, security telemetry, and system log parsing.
- Long-Context Applications: Suitable for scenarios demanding extensive context tracking without degradation, such as analyzing large codebases or detailed technical documentation.
- Uncensored Technical Debate: Provides direct, unfiltered technical responses without policy-based refusals.