Cisco's open-weight Antares model localizes vulnerabilities, rivals models 100x larger
aminkarbasi · x · 2026-09-30
Cisco chief AI scientist Amin Karbasi introduced Antares, an open-weight small model built specifically for vulnerability localization. It narrows thousands of files to a ranked set of likely culprits, runs locally to keep source code private, and matched or beat models 100x larger in internal and independent benchmarks. Open weights target universities, nonprofits, and public-sector teams.
More from Models
- GPT-6.1-Sol shifts the intelligence-per-dollar frontier, but Opus 5.5 still tops raw scores — randal_olson · 2026-09-30
- User claims $500 ChatGPT plan hit weekly limit in 30 minutes, unverified — RexDouglass · 2026-09-30
- ChatGPT, Claude and Grok Suffer Partial Outages at the Same Time — rohanpaul_ai · 2026-09-30
- OpenAI's $500 DevDay plan mocked as model falls behind Claude Opus 5.5 — XFreeze · 2026-09-30
- OpenAI's $500 plan burns through all limits in 40 minutes on ultrafast mode — imjustnewatai · 2026-09-30
- Linear extrapolation puts rumored GPT-6.1 Astra at ~79.5-80% on OSWorld 2.0 — ChrisGPT · 2026-09-30