OrcaRouter, a prominent player in the AI landscape, has unveiled OrcaCyber Zero 1.5, an advanced artificial intelligence model tailored for authorized vulnerability research. This latest iteration succeeds OrcaCyber Zero 1.0, which debuted in late 2026. Designed for security professionals, OrcaCyber Zero 1.5 aims to streamline processes like vulnerability reproduction, exploit creation, and penetration testing. It integrates a substantial 1-million-token context window, along with built-in function execution capabilities and structured data outputs. The overarching message for security teams is a focus on fewer initial reports and an increase in validated and resolved security flaws.
Core Capabilities and Design Principles
OrcaCyber Zero 1.5 represents a significant advancement in frontier cybersecurity models, crafted specifically for security engineering and research applications. Its core uses encompass security auditing, cyber reasoning, and the aforementioned vulnerability tasks. The model was developed with three primary objectives:
- Identifying Overlooked Flaws: To uncover complex, previously undetected vulnerabilities such as remote code execution, sandbox escapes, authentication bypasses, privilege escalations, and intricate attack chains.
- Beyond Simple Detection: To analyze full attack paths, critically evaluate its own hypotheses, and prioritize security issues based on their demonstrable exploitability.
- Empowering Autonomous Agents: Featuring its extensive context window and native tool interactions, the model is engineered to support advanced reasoning across vast codebases, making it ideal for autonomous security systems.
Performance Benchmarks
The model's official page presents several performance metrics, last updated in October 2026. These figures, reported by the vendor, position OrcaCyber Zero 1.5 favorably in specialized cybersecurity challenges:
- Cybench: A perfect 100% score across 39 out of 40 professional Capture The Flag (CTF) tasks, demonstrating unrestricted agent execution.
- CVE-Bench: Achieved 95.8% on a subset of 24 evaluable tasks derived from 40 critical-severity web Common Vulnerabilities and Exposures (CVEs).
- HumanEval+: Recorded a strong 93.9% performance.
- SWE-bench Pro V2: Its lowest disclosed score stands at 76.5%, though this particular benchmark's results are not directly comparable with standard SWE-bench Pro outcomes.
It is important to note that all reported scores originate from the vendor, and independent verifications are not yet available.
Access and Pricing Structure
Access to OrcaCyber Zero 1.5 is managed through a hosted API, provided exclusively by OrcaRouter. The model's underlying weights and hardware specifications have not been publicly disclosed. Developers can interact with the model via an OpenAI-compatible API, setting the base URL to https://api.orcarouter.ai/v1 and targeting the orca/orcacyber-zero-1.5 endpoint.
Entry to the model is restricted to a dedicated Security Research tier. This tier requires a specific engagement, a valid passkey, and acceptance of predefined terms, targeting trusted security researchers, red teams, and entities conducting authorized testing.
The pricing model for the service is set at $3.00 per 1 million input tokens and $7.50 per 1 million output tokens. Additionally, cache reads incur a charge of $0.75 per 1 million tokens. Recent latency figures, based on a limited traffic sample of 1,300 tokens, indicated a p50 time-to-first-token of 500 milliseconds and a p95 of 2.36 seconds. These figures are not directly comparable to OrcaCyber Zero 1.0’s recorded 3.43 seconds p50, which was measured over a significantly larger traffic volume.
Competitive Landscape and Outlook
While direct comparisons are complex due to varying evaluation methodologies and non-public data from competitors, OrcaCyber Zero 1.5's pricing structure notably undercuts some existing advanced models. For instance, its token pricing is considerably lower than that of Claude Mythos Preview. The model represents a strategic offering from OrcaRouter, aiming to provide powerful, specialized AI capabilities to the cybersecurity community.
The information provided by OrcaRouter emphasizes the model's high performance in cyber-specific benchmarks and its potential to enhance the efficiency and effectiveness of security operations. The lack of independent reports means the community will likely await further evaluations to fully ascertain its real-world impact and compare it rigorously against other solutions.
This article is a rewritten summary based on publicly available reporting. For the original story, visit the source.
Source: MarkTechPost
