Emergency open-weight release expands the OrcaCyber Nano, Zero, and Ultra family, combining private local inference with frontier-scale cybersecurity capabilities.
SAN FRANCISCO, Oct. 20, 2026 /PRNewswire/ -- OrcaRouter today announced the emergency release of OrcaCyber Nano 1.0, a 27-billion-parameter open-weight cybersecurity model designed to bring advanced coding, reasoning, and defensive security capabilities to consumer-grade hardware.
Nano expands the OrcaCyber model series — Nano, Zero, and Ultra — offering specialized AI for cybersecurity workloads across local devices, enterprise environments, and frontier-scale security research.
Frontier-Class Coding Performance in 27B Parameters
OrcaCyber Nano 1.0 reports strong performance across software engineering, coding, and reasoning benchmarks:
Nano's reported 78.4% on SWE-bench Verified approaches DeepSeek V4 Flash's reported 79.0%, a difference of 0.6 percentage points. The evaluations may use different configurations and are not a controlled head-to-head comparison.
The model supports a 262K-token context window, vision capabilities, and offline inference. A quantized configuration is designed for local deployment within 16GB of GPU memory. This enables security teams to work with sensitive source code, binaries, and logs without sending them to external inference services.
OrcaCyber Zero 1.5: Advanced Autonomous Security Research
Alongside Nano, OrcaRouter offers OrcaCyber Zero 1.5, a specialized cybersecurity model for vulnerability research, authorized security testing, auditing, and multistep security reasoning. Reported model evaluations include:
Separately, the OrcaCyber autonomous research harness reported 98.01% pass@1 on CyberGym Level 1, reproducing 1,477 of 1,507 vulnerabilities across 188 open-source projects. This is a combined agent-system result using OrcaCyber Zero 1.0, DeepSeek V4 Flash, and OrcaRouter adaptive routing; it is not a standalone Zero 1.5 score.
OrcaCyber Zero 1.5: https://www.orcarouter.ai/models/orca/orcacyber-zero-1.5
One Cybersecurity Model Family, Multiple Deployment Scales
Nano: Compact, open-weight models for private local inference, vulnerability analysis, and defensive code review.
Zero: Specialized frontier models for authorized vulnerability research, automated security testing, and incident response.
Ultra: Models aimed at complex cybersecurity reasoning and advanced autonomous security workflows.
OrcaRouter plans additional CyberGym and security-specific evaluations across the series. Access to sensitive cybersecurity capabilities may require verification.
Availability
OrcaCyber Nano 1.0 (open weights): https://huggingface.co/orcarouter/OrcaCyber-Nano-1.0
OrcaCyber Zero 1.5: https://www.orcarouter.ai/models/orca/orcacyber-zero-1.5
Explore all OrcaCyber models and request access: https://www.orcarouter.ai/models?tier=cyber
About OrcaRouter
OrcaRouter is an AI infrastructure platform offering unified access to models and tools through one API, with intelligent routing, observability, security controls, and flexible deployment options. Its OrcaCyber family focuses on AI for cybersecurity research, software security, and autonomous defense.
One Gateway. Every Model. | https://www.orcarouter.ai
SOURCE OrcaRouter
Read More : How AI-Driven Threat Detection Stops Modern Cyber Attacks in Real Time?
