Cisco Foundation AI has introduced Antares, a new family of open-weight small language models (SLMs) designed specifically for vulnerability localization in code. The Antares models, available in 350M and 1B parameter sizes, are built upon IBM Granite 4.0 checkpoints and are released under the Apache 2.0 license on Hugging Face. Alongside the models, Cisco has also released the Vulnerability Localization Benchmark (VLoc Bench), a 500-task evaluation suite derived from real-world GitHub Security Advisories. While Antares-3B performs comparably to GPT-5.5 on this specific task, the 1B model demonstrates superior performance to larger open-weight models like GLM-5.2, highlighting the effectiveness of task-specific training over sheer parameter count. AI
IMPACT Sets new SOTA on vulnerability localization benchmarks, potentially improving code security tooling.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Antares
- Antares-1B
- Antares-350M
- Apache Software License 2.0
- Cisco Foundation AI
- Common Weakness Enumeration
- Gemini 2.5 Flash
- Gemma-4-31B
- GitHub Security Advisories
- GLM-5.2
- GPT-5.5
- Hugging Face
- IBM Granite 4.0
- VLoc Bench
- Vulnerability Localization Benchmark
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
