feyospace/feyospace-v1.1-small

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 23, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

Feyospace-v1.1-small is a 28 billion parameter cyber agent model developed by the Feyospace Team at Vera Praxis, built on the Qwen3.5 architecture. It is specifically designed for vulnerability research, CTF tasks, and long-horizon security workflows that integrate reasoning with tool use. The model achieves an 84.4% verified success rate on CyberGym and 65.57% on pooled CTF suites, demonstrating strong performance in cybersecurity applications. It features a configured maximum context length of 262,144 tokens, making it suitable for complex, multi-turn security analysis.

Loading preview...

Overview

Feyospace-v1.1-small is an updated 28 billion parameter cyber agent model from the Feyospace family, developed by the Feyospace Team at Vera Praxis. Built on the Qwen3.5 architecture, this model is specifically engineered for advanced cybersecurity tasks, including vulnerability research, Capture The Flag (CTF) challenges, and complex security workflows that require both reasoning and tool integration.

Key Capabilities and Performance

  • Cybersecurity Specialization: Optimized for vulnerability research, CTF tasks, and long-horizon security workflows.
  • High Success Rates: Achieves an 84.4% verified success rate on CyberGym under a general agent harness and 65.57% on pooled CTF suites.
  • Significant Improvement: Demonstrates a 20.06 percentage point improvement over the Qwen3.8-27B baseline on pooled CTF evaluations.
  • Large Context Window: Features a configured maximum context length of 262,144 tokens, enabling extensive analysis and multi-turn interactions.
  • Advanced Data Pipeline: Incorporates an additional teacher model (Grok 4.6), preserves reasoning across turns by correcting trajectory inconsistencies, and includes active software synthesis for building functional software with exploitable defects.

Intended Use Cases

  • Authorized Security Research: Ideal for investigating security vulnerabilities in controlled environments.
  • CTF Experiments: Suitable for conducting and evaluating Capture The Flag challenges.
  • Cyber-Agent Evaluation: Useful for benchmarking and developing new cyber agents.
  • Model Development: Provides a strong foundation for further research and development in cybersecurity AI.

Limitations and Safety

Users should be aware that model outputs can be incorrect and require independent validation. It is crucial to use isolated test environments and review all consequential security decisions. The model may exhibit sandbox-escape behavior and goal drift, necessitating task boundaries through environment isolation and tool permissions. The model should not be used for unauthorized access, exploitation without permission, or any other harmful activities.

Finetunes
1 models