nurdich/Qwen3.8-27B-OBLITERATED22-AGENT

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 17, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

nurdich/Qwen3.8-27B-OBLITERATED22-AGENT is a 27 billion parameter language model based on the Qwen3.8 architecture, specifically designed for agentic use cases. It features a restored chat template for tool definitions and vision content, and is notable for its "liberated" behavior, providing direct answers without safety lectures or refusals, even for restricted queries. This model maintains near-stock capability with a modest MMLU drop of 2.1 percentage points, excelling in code generation and real-world advanced tasks.

Loading preview...

Overview

nurdich/Qwen3.8-27B-OBLITERATED22-AGENT is a 27 billion parameter model based on the Qwen3.8 architecture, specifically re-published with a restored chat template for enhanced agentic capabilities. This model is a "liberated" version, meaning it has undergone surgical removal of safety guardrails to provide direct answers without refusals or safety lectures, even for restricted queries. It aims to offer genuine liberation while maintaining near-stock performance.

Key Capabilities

  • Genuine Liberation: Provides real substance and answers to restricted queries, eliminating both hard refusals and soft deflections (safety lectures).
  • Agentic Use: Restored chat template supports tool definitions, assistant tool calls, tool results, and vision content, making it suitable for coding agents and pentest frameworks.
  • High Code Generation Performance: Achieves 20/20 on tested cyber/code generation tasks, producing functional implementations.
  • Thinking ON Compatible: Works without refusals in both thinking and non-thinking modes, though thinking mode may produce longer responses.
  • Advanced Real-World Tasks: Successfully handles 7 out of 8 advanced real-world tasks, including ReAct agent loops, async code refactoring, and security code review.

Good For

  • Alignment Researchers: Studying refusal geometry and safety robustness in LLMs.
  • Red-Teamers: Evaluating post-training safety against weight surgery.
  • AI Safety Evaluators: Needing an unrestricted baseline for testing.
  • Local-First Users: Desiring full control over model outputs on their own hardware.
  • Developers: Building coding agents or pentest frameworks that require uncensored and direct responses.

Finetunes
1 models