Cisco’s Antares-1B Model Enhances Vulnerability Localization

• •Antares-1B is a new 1-billion parameter model for vulnerability localization.
• •The model uses a terminal interface to autonomously navigate and analyze codebases.
• •Antares-1B outperforms larger models in vulnerability detection efficiency.

Cisco has released Antares-1B, a language model designed for vulnerability localization in codebases. This model, part of the Antares family, utilizes a two-stage training pipeline combining supervised fine-tuning and reinforcement learning. Antares-1B operates as a terminal agent, executing commands to explore code repositories and identify vulnerabilities. It achieves a File F1 score of 0.209 on the VLoc Bench, outperforming larger models like GLM-5.2. The Antares family includes models with 350M, 1B, and 3B parameters, all based on IBM Granite. The 1B model is noted for its efficiency, completing extensive evaluations quickly and at a low cost. This release marks a significant advancement in automated vulnerability detection tools.