The Agent RaceJuly 22, 2026via MarkTechPost
Cisco Foundation AI Releases Antares: 350M and 1B Open-Weight Models That Localize Known Vulnerabilities Inside Real Codebases
Why it matters
Cisco's open-weight models demonstrate that specialized, smaller models can outperform frontier models on narrow security tasks at a fraction of the cost—reshaping economics for enterprise security tooling.
Key signals
- Antares family: 350M and 1B parameter models
- Antares-1B achieves 0.209 File F1 on Vulnerability Localization Benchmark
- Outperforms GLM-5.2 (753B parameters) and Gemini 3 Pro
- 500-task sweep: ~13 minutes on single H100, <$1 cost
- GPT-5.5 cost: $141 per 500-task sweep
- Post-training critical: untrained Granite 4.0 baseline scores near zero
- Open-weight release enables enterprise deployment without API dependency
The hook
Antares-1B beats Gemini 3 Pro on vulnerability detection. $1 per task vs. $141 for GPT-5.5.
Cisco Foundation AI has released Antares, a family of small language models trained to pinpoint where known vulnerabilities live inside a codebase. Antares-1B reaches 0.209 File F1 on the new Vulnerability Localization Benchmark, above GLM-5.2 at 753B parameters and Gemini 3 Pro. The untrained Granite 4.0 checkpoints score near zero under the same protocol, so post-training supplies almost all of the capability. A full 500-task sweep runs in roughly 13 minutes on a single H100 for under a dollar, against $141 for GPT-5.5.
The post Cisco Foundation AI Releases Antares: 350M and 1B Open-Weight Models That Localize Known Vulnerabilities Inside Real Codebases appeared first on MarkTechPost.