Cisco Foundation AI Releases Antares: 350M and 1B Open-Weight Models for Vulnerability Localization in Codebases
Cisco Foundation AI has released Antares, a family of small language models designed to identify the locations of known vulnerabilities within codebases. Antares-1B achieves a 0.209 File F1 score on the new Vulnerability Localization Benchmark, outperforming much larger models such as GLM-5.2 (753B parameters) and Gemini 3 Pro. Running a full 500-task benchmark sweep takes about 13 minutes on a single H100 GPU for under a dollar, compared to $141 for GPT-5.5.
Why it matters: Antares shows that small, specialized models can surpass much larger general-purpose models in targeted security tasks, offering a more cost-effective solution for vulnerability localization.
Full story at: MarkTechPost / AI ↗