OpenAI previews Astra: a cybersecurity model scoring 100% on ExploitBench
LingmingZhang · x · 2026-09-02
OpenAI has unveiled 'Astra,' a cybersecurity model that achieved a perfect 100% score on the public ExploitBench benchmark. In internal, contamination-free tests based on post-cutoff vulnerabilities, Astra demonstrated 4x the cyber capability of GPT-4.06 (5.6) using fewer tokens. The team emphasized significant efforts to align the model's safeguards with its advanced capabilities, meeting the 'Critical' threshold under OpenAI's Preparedness Framework.
More from Models
- Mocked benchmark shows DeepSeek's document chart grounding badly misaligned — burkov · 2026-10-18
- Dev complains Claude's habit of saying "not done" is his biggest pain point — jasonkneen · 2026-10-18
- Open-weight TYGER speech decision model makes 80ms turn-taking calls for full-duplex voice agents — solyarisoftware · 2026-10-18
- Paying Google AI subscriber blocked: custom-domain Workspace accounts can't access AI Studio — ChrisRackauckas · 2026-10-18
- Running Ling 3.1 Flash (560B MoE, 25B Active) Across 4 DGX Sparks via llama.cpp RPC — BerryBlush_Rogers691 · 2026-10-18
- Users press OpenAI on long-delayed Adult Mode a year after it was promised — ___Patrice___ · 2026-10-18