Singapore-based CONTINUUM AI PTE. LTD., through its OpenAI-compatible LLM gateway OrcaRouter, has introduced OrcaCyber Nano 1.0, an open weight model designed for cyber defense. The company described the launch as urgent because defenders cannot wait for delayed evaluations. It published initial benchmark results and promised cyber-specific evaluations later, signaling a focus on immediate operational use.
Why the Model Was Released Early
OrcaRouter stated that the release was accelerated because security teams cannot pause operations while waiting for dedicated cyber evaluations. The company framed the announcement as an emergency release rather than a standard model launch. It explained that cyber-specific assessments will follow, but the model is already available to support active defensive work.
Technical Specifications and Offline Operation
OrcaCyber Nano 1.0 is a 27 billion parameter dense model with a 262K context window, vision input support, and fully offline operation. OrcaRouter said it runs locally under 16GB, which makes it suitable for restricted environments that cannot send data outside. The company claims it is the highest performing model in the under 16GB local category based on its own initial evaluations.
Benchmark Results and SWE-bench Verified Score
The model card lists six results: GSM8K 97.50%, MATH-500 98.60%, AIME 2025 96.23%, MBPP+ 84.66%, IFEval 82.81%, and SWE-bench Verified 78.4%. The first five were included as initial benchmarks, while SWE-bench Verified was added after full results arrived. OrcaRouter noted the SWE-bench figure approaches the 79.0% reported for DeepSeek V4 Flash.
Intended Use Cases and Product Family
OrcaRouter said the model is built for AI defense, vulnerability research, code security, and incident response. The company also suggested using models from the same family through a single API and wallet, allocating workloads based on cost efficiency. The broader OrcaCyber series includes Nano, Zero, and Ultra, with Zero 1.5 available via API for teams that prefer managed access.
Local Deployment and Model Access
Users can download the model through the command huggingface-cli download orcarouter/OrcaCyber-Nano-1.0 from the listed repository. OrcaRouter emphasized complete offline functionality for defensive operations in environments that must remain isolated. The company positions Nano 1.0 as the local option, while other models in the series can be accessed through its API, giving teams a choice between local control and managed convenience.
Gateway Portfolio and Cost Efficiency
OrcaRouter provides access to more than 200 models through one OpenAI-compatible endpoint, including OrcaCyber Zero 1.5, OrcaCyber Ultra 1.0, DeepSeek V4.1 Flash, Claude Opus 5.5, GPT-6.1 Sol, Qwen3.8 Max, and GLM-5.3 Flash. The platform offers roughly 40% cost savings, routing overhead below 1 millisecond, and no token markup. Its self-hosted edition, OrcaRouter-Lite, is available under an MIT license.
About OrcaRouter and CONTINUUM AI
CONTINUUM AI PTE. LTD. is a Singapore-based company founded to make advanced AI accessible and affordable for developers and teams. OrcaRouter is its OpenAI-compatible AI Gateway, providing a single API key and endpoint for more than 200 leading models. The company also offers OrcaRouter-Lite under an MIT license for teams that want a self-hosted routing option.
CONTINUUM AI continues to position OrcaRouter as a practical gateway for developers and security teams that need flexible model routing. The early release of OrcaCyber Nano 1.0 signals a focus on defensive cyber operations where latency, data isolation, and offline access matter. With local deployment, competitive benchmark results, and integration into a larger model catalog, the announcement gives security practitioners another open weight option to evaluate.