OpenAI's Astra scores 100% on ExploitBench, discovers zero-days in internal test
zephyr_z9 · x · 2026-09-02
OpenAI's new model Astra demonstrates a significant leap in cybersecurity capabilities. It scored 100% on the ExploitBench benchmark, surpassing Sol-5.6 (Max) at 73.5% and Mythos 5 at 78%. To prevent benchmark contamination, OpenAI tested it on a new internal benchmark with recent vulnerabilities. With roughly comparable output tokens (77k), Astra scored 39% while GPT-5.6 Sol scored only 1%. During evaluation, Astra also discovered and utilized two zero-day vulnerabilities as part of an exploit chain.
More from Models
- Mocked benchmark shows DeepSeek's document chart grounding badly misaligned — burkov · 2026-10-18
- 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