OrcaRouter Releases OrcaCyber Zero 1.5 Cybersecurity Model

OrcaRouter has released OrcaCyber Zero 1.5, a new cybersecurity model designed for authorized vulnerability research, exploit development, and penetration testing. This model is the successor to OrcaCyber Zero 1.0, which was initially released on September 17, 2026. OrcaCyber Zero 1.5 is a post-trained Orca model specifically engineered to aid security professionals in reproducing vulnerabilities, developing exploits, and conducting penetration tests. A key feature of OrcaCyber Zero 1.5 is its substantial 1 million token context window, which allows for the processing of extensive amounts of information. The model also incorporates native function calling capabilities and provides structured outputs, aiming to streamline the reporting process for security teams by enabling them to focus on validating and fixing vulnerabilities rather than generating lengthy reports. The parameter count for OrcaCyber Zero 1.5 has not been disclosed by OrcaRouter. The model supports text input and text output, with a maximum output length of 128,000 tokens. Access to OrcaCyber Zero 1.5 is exclusively provided through a hosted API offered by OrcaRouter; the model weights and any quantized variants are not available, and the specific hardware it runs on has not been disclosed. OrcaRouter reports that OrcaCyber Zero 1.5 achieves near-ceiling scores on cybersecurity benchmarks and demonstrates strong performance in coding tasks. Specifically, the model achieved a perfect score of 100% on Cybench, successfully completing all 39 tasks with unrestricted agent execution. Its lowest published score was 76.5% on SWE-bench Pro V2. OrcaRouter highlights the model's top-tier cybersecurity performance at a cost of $3.00 per 1 million tokens for input and $7.50 per 1 million tokens for output. However, the company acknowledges that all performance numbers are self-reported and that the CVE-Bench evaluation used only 24 evaluable tasks. The three primary design goals for OrcaCyber Zero 1.5 are to discover unknown flaws such as RCE, sandbox escapes, authentication bypasses, privilege escalation, and attack chains; to move beyond simple detection by reasoning through attack paths, challenging its own hypotheses, and ranking flaws by exploitability; and to be built for autonomous agents, leveraging its 1 million token context, native tool calling, and extended reasoning for large codebases. The model page lists four vendor-reported performance results, with the last evaluation conducted on October 10, 2026. The model's capabilities are intended to assist security teams in finding vulnerabilities that might be missed by other tools and in understanding the practical exploitability of identified security weaknesses.