Breaking

Community Releases Uncensored Mythos-Class Build of Qwen3.8-27B for Local Agents

September 24, 2026 at 23:46 EDT

Foundation Models
AI Agents
Security

A community release has turned Qwen3.8-27B into an uncensored, agent-oriented build that pairs refusal-removal work with a new planning and tool-calling template. The model, medismera/Qwen3.8-27B-OBLITERATED-Mythos-Class-Agentic, is listed as an Apache-2.0 derivative of OBLITERATUS/Qwen3.8-27B-OBLITERATED and Qwen's open 27B vision-language backbone.

Sep. 25, 2026 · Qwen3.8-27B

Uncensored agent build reaches local hardware

A Mythos-Class derivative pairs refusal-removal work with long-context planning, tool calling and quantized files sized for local inference.

27B

base parameters

128K

production context

11.0GB

smallest GGUF listed

reported refusal rate

Context scale-up

The production context is four times the upstream 32K stub described by the Mythos-Class card.

32K

upstream stub

128K

production setting

What changed in the agent stack

Abliterated weights

→

Task-tree planning

→

Tool-call routing

Why builders care

Private coding agents, local security testing and long-context automation become more practical when the files fit desktop inference stacks.

Why it is sensitive

The same edits intentionally weaken refusals, and testing shows truncation can make safety results easy to misread.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.