Leaked slide shows OpenAI's unannounced cyber model hitting 77% on ExploitGym benchmark
TheMoonMidas · x · 2026-11-20
A photo of an internal OpenAI display leaked ExploitGym (exploit benchmark) results: an unannounced 'Upcoming Cyber Model' reaches 77.0% success, versus 42.4% for GPT-6 Astra and 52.6% for GPT-6 Cyber. The chart also shows success climbing with average output tokens up to 80K, suggesting heavier inference compute. Unconfirmed leak.
More from Models
- Local inference users want 512GB RAM and 1M context, says current boxes fall short — AIFlow_ML · 2026-11-20
- Astra xhigh is surprisingly good at ML engineering, says tester — _Stocko_ · 2026-11-20
- Beth Barnes clarifies METR didn't evaluate alignment of Fable/Mythos 5.2 — RyanGreenblatt · 2026-11-20
- Tencent Hy4 lands in vLLM day-0: 774B MoE with 49B active, 1M context, 8-GPU FP8 recipe — vllm_project · 2026-11-20
- Google Now Ships Two Models a Week, Six of Top 10 Spots on Artificial Analysis Index — haider1 · 2026-11-20
- Developer Says Gemini Neon Is His Favorite All-Purpose Model for Technical Writing and Slides — BenBlaiszik · 2026-11-20