Leak: OpenAI's internal cyber model scores 77.0% on ExploitGym, near-doubling GPT-6 Cyber's 52.6%
haider1 · x · 2026-11-13
Unverified leak circulating on X: OpenAI's internal cyber model reportedly hit 77.0% success on ExploitGym, nearly double GPT-6 Cyber's 52.6% and well above GPT-6 Astra's 42.4%. The shared chart shows an "Upcoming Cyber Model" curve still climbing toward benchmark saturation. The poster speculates this is a GPT-6.5 Cyber, distilled via specialized post-training from OpenAI's large internal Astra pre-trains, and argues near-saturating a cyber benchmark signals rapid general capability progress. Not officially confirmed.
More from Models
- Why AI Slop Is So Frustrating to Read: Hidden Subjects and Buried Actions — teortaxesTex · 2026-11-13
- MSR: Rewriting fewer than 10 dormant neurons backdoors any LLM agent, no training needed — mikeflache · 2026-11-13
- Google's Logan Kilpatrick asks users what to ship next; community wants a new Gemma — jacek2023 · 2026-11-13
- Microsoft diagnostic AI solves 83% of NEJM cases vs 36% for physicians — realmeetjames · 2026-11-13
- Opus 5.5 generates a full Blender 3D model in seconds, wows game developer — Scobleizer · 2026-11-13
- Claude's JavaScript art demo draws praise from threepointone: 'just remarkable' — threepointone · 2026-11-13