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.
Related event: OpenAI's Astra Scores 100% on ExploitBench Cybersecurity Benchmark(7 posts)→
More from Models
- Screenshots suggest Gemini 2.0 Flash 3.8 is rolling out — Maglcite · 2026-09-02
- Fable 5.1 Still Struggles with Estimates, User Reports — rudrank · 2026-09-02
- Fable 5.1 recreates Mario World 1-1; critics call for a level-design benchmark — max_paperclips · 2026-09-02
- 3B TwIL Model Outperforms 120B Open Source Model on Formal Reasoning — Socially-great8275 · 2026-09-02
- Model Scaling Trend: 2T Parameters Becoming New Norm as KV Cache Shrinks 10x YoY — zephyr_z9 · 2026-09-02
- V4-Flash-Vision-Exp Outperforms GLM 5.3 in Coding Benchmark — teortaxesTex · 2026-09-02