OpenAI demos unreleased cyber model scoring 77% on ExploitGym, 24 points above shipped GPT-6 Cyber
chetaslua · x · 2026-11-13
At its 'Cyber at the Frontier' keynote, OpenAI showed an unreleased cyber model scoring 77% on ExploitGym exploit success, well ahead of Gemini 4 Carbon (54.8%), GPT-6 Cyber (52.6%, shipped just 8 days earlier), GPT-6 Astra (42.4%), Claude Mythos 5.2 (38.1%) and Claude Opus 5.5 (33.3%).
On PatchBench it also leads with 79.1% vs 73.5% (GPT-6 Cyber) and 71.1% (GPT-6 Astra). Note: all vendor-reported scores.
Related event: OpenAI's unreleased Cyber model scores 77% on ExploitGym(12 posts)→
More from Models
- Rumored Gemini 4 Carbon reportedly dropping today, billed as 'best model in the world' — zacharynado · 2026-11-14
- Google's packed week: Gemini 4 Carbon, Decisions API, Robotics 2.5, AlphaGenome 2 — osanseviero · 2026-11-14
- Dwarkesh: The avalanche hitting math is coming for AI development — Dwarkesh Podcast · 2026-11-14
- Redditor: Gemini Hydrogen summarizes YouTube videos 30x faster, in under 1 second — TaurusGold · 2026-11-14
- Anthropic report: scientists across fields use Claude to write, not to code — RexDouglass · 2026-11-14
- 'Quietly' is the new 'delve': mocking LLM writing tics — sebkrier · 2026-11-14