OpenAI Announces New Artificial Intelligence Model GPT-6 Astra
Announced by OpenAI, the GPT-6 Astra model achieved one hundred percent success in the ExploitBench test, showcasing its advanced cybersecurity capabilities.
OpenAI has officially announced GPT-6 Astra, described as a new turning point in the field of artificial intelligence and its smartest model to date. Drawing attention with its advanced cybersecurity and software engineering capabilities, this new model initially prevents the generation of exploit codes as part of security measures.
GPT-6 Astra Unveiled
In an announcement on Thursday, OpenAI officially unveiled GPT-6 Astra, stating that it is the smartest and most adaptable artificial intelligence model worldwide. This development comes just days after it was announced that the model crossed the Critical cybersecurity capability threshold under its Preparedness Framework.
Record Success in Tests
The Astra model has reached a state-of-the-art level in computer use, browsing, software engineering, cybersecurity, science, and professional business fields. The model achieved a score of one hundred percent on the ExploitBench test, ninety-eight percent on the FrontierMath Tier 4 test, and ninety-nine point nine percent on the ARC-AGI-3 test.
Accessibility and Platforms
The new artificial intelligence model has already begun rolling out to a small group of organizations. The model is expected to reach a broad audience in the upcoming period through ChatGPT Plus, Pro, Business, and Enterprise users, as well as the OpenAI API, Microsoft Azure, and Amazon Web Services Bedrock.
Cybersecurity and ExploitBench Performance
On ExploitBench, which evaluates the ability to turn known software vulnerabilities into functioning exploits, Astra achieved flawless success, surpassing the seventy-eight point five percent score of its predecessor, GPT-5.6 Sol.
Security Measures and Restrictions
Considering the dual-use nature of these capabilities, the released version of Astra has been limited to secure code review and patch generation. The system refuses to fulfill prompt requests related to generating proof-of-concept exploits.
Daybreak Initiative
OpenAI also launched the one billion dollar Daybreak for Frontline Defenders initiative to provide subsidized access, training, and technical assistance to critical infrastructure sectors. A joint pilot program is also being conducted with the US Multi-State Information Sharing and Analysis Center under this scope.