Cisco Foundation AI releases Antares vulnerability localization models
Cisco Foundation AI released Antares, a family of security small language models designed to localize vulnerabilities in codebases. Two open-weight models are available on Hugging Face under Apache 2.0 license: Antares-350M and Antares-1B. Cisco also released the Vulnerability Localization Benchmark (VLoc Bench), a 500-task agentic evaluation under the same license, drawn from 290 unique real-world repositories across six ecosystems: npm, pip, Maven, Go, Rust, and Composer. The Antares-1B model achieves a File F1 score of 0.209, compared to GPT-5.5 at 0.229 and a 753B open-weight model at 0.186. All three Antares models initialize from IBM Granite 4.0 checkpoints and use decoder-only transformer architecture with grouped-query attention, SwiGLU MLPs, RMSNorm, RoPE, and shared input/output embeddings. The models operate within a constrained agent loop with access to read-only terminal commands in a sandboxed environment, with a budget of 15 terminal calls per task.