OpenAI Astra achieves 3.4x exploit success rate vs GPT-5.6 with 45% fewer tokens
imjustnewatai · x · 2026-09-02
OpenAI's upcoming security model Astra shows strong performance in internal testing. On a set of 20 high-severity vulnerabilities, Astra achieved a 39% exploit success rate using 76k output tokens, compared to GPT-5.6's 11.5% using 138k tokens. This represents a 3.4x higher success rate with 45% fewer tokens. Astra also discovered and used two zero-days in an exploit chain. While Fable 5.1's launch sparked comparisons, Astra is not yet public. The real shift is cheaper repetition (more exploit paths per token), with OpenAI countering via runtime supervision.
More from Models
- 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
- Polymarket opens betting market on who releases the next Gemini Argon model — Polymarket · 2026-10-18