GPT-6 Cyber Sets Record in Cybersecurity – 48.5% ExploitGym Score
GPT-6 Cyber achieves a 48.5% ExploitGym score, finds 81 new Firefox vulnerabilities and 948 Linux bugs, now available via Codex and API for Daybreak Red participants.
Best model in the world for cybersecurity tasks is here!
Dear ladies and gentlemen, meet GPT-6 Cyber.
According to published results, the model sets a new record in cybersecurity tasks. It scored 48.5 % on the ExploitGym benchmark, discovered 81 previously unknown high‑severity vulnerabilities in Firefox, and showed how 948 bugs in Linux can be exploited by attackers to expand their capabilities.
Currently GPT-6 Cyber is available in Codex and via API only to selected participants of the Daybreak Red program.
And the expected GPT-6.2 Astra and Sol have not been released today. Now we await tomorrow.
Comments
Weekly digest
The best of vibe coding, AI agents and open source — once a week, no spam.
Related articles
All articlesOpenAI Frontier Models and Codex Now Available on AWS via Bedrock
OpenAI’s frontier models and Codex are now on AWS through Amazon Bedrock, enabling secure, compliant adoption with existing AWS governance and future integrations.
OpenAI Tests New Model by Rewriting Codex in Rust for Cross‑Platform
OpenAI’s Andrew Ambrosino says each new model must rewrite the desktop Codex in Rust and GPUI, aiming for a single model that can build native Mac, Windows and cross‑platform apps.
OpenAI Launches GPT-6 Sol and Luna: Cheaper, Efficient AI Models
OpenAI's new GPT-6 Sol and Luna are twice as cheap as their 5.6 versions, offering clear communication and cost‑effective performance for your AI needs.