MaanVad3r/Antanom

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 6, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Antanom is a 27 billion parameter Qwen-based model developed by MaanVad3r, specifically engineered for controlled cybersecurity research and authorized security testing. This model is designed to reduce refusal behavior on cybersecurity prompts while retaining the base model's general capabilities. It features bfloat16 weights and supports text and vision inputs, with a maximum context length of 262,144 tokens. Its primary differentiator is its focus on enabling cybersecurity research by modifying refusal patterns.

Loading preview...

Antanom: A Research-Oriented Cybersecurity Model

Antanom is a 27 billion parameter model created by MaanVad3r, derived from a merged and abliterated version of Qwen/Qwen3.8-27B. Its core purpose is to facilitate controlled cybersecurity research and authorized security testing by intentionally altering refusal behaviors for cybersecurity-related prompts.

Key Capabilities and Features

  • Reduced Refusal on Cybersecurity Prompts: Engineered to provide responses to cybersecurity queries that other models might refuse, enabling deeper research into security topics.
  • Base Model Capabilities Retained: While specialized for cybersecurity, it aims to maintain the general language understanding and generation capabilities of its Qwen base.
  • Technical Specifications: A 27B-class Qwen model with 64 transformer layers, utilizing bfloat16 weights. It supports both text and vision inputs.
  • Extended Context Length: Configured for a maximum position setting of 262,144 tokens.
  • Development Method: Created using Heretic directional low-rank weight editing followed by a full model merge, rather than a PEFT adapter.

Intended Use Cases

Antanom is specifically designed for:

  • Offline evaluation and synthetic cybersecurity benchmarks.
  • Secure-code education and incident-response exercises.
  • Testing in isolated systems with explicit operator authorization.

Important Limitations and Risks

It is crucial to understand that Antanom's refusal behavior is intentionally modified for research, meaning ordinary safety expectations do not apply. The model can produce incorrect, incomplete, or unsafe output. Users are responsible for ensuring compliance with laws and organizational rules, and it should never be used to target real systems, obtain secrets, or perform unauthorized activities. The evaluation snapshot is small and task-specific, not establishing reliability or general safety for production deployment.

Finetunes
1 models