SAN FRANCISCO (TechGenez) – OpenAI has introduced its most powerful AI model yet, GPT-6 Astra, describing it as the world’s most intelligent and aligned system to date.
The new model brings together years of research across pre-training, reinforcement learning and alignment. It is now available to a limited group of organisations and will soon be rolled out to all ChatGPT Plus, Pro, Business and Enterprise users, as well as through the OpenAI API, Microsoft Azure and AWS Bedrock.
OpenAI says GPT-6 Astra sets new benchmarks in computer use, browsing, software engineering, cybersecurity, science and professional work. It also scores a perfect 100% on ExploitBench and achieves a 98% score on FrontierMath Tier 4, having already helped solve long-standing open problems in mathematics.
Strong Results in Major Benchmarks
The model performed particularly well in tests designed to measure real-world problem-solving:
- On Terminal-Bench, a test for AI agents in terminal environments, GPT-6 Astra scored 87.9, nearly matching the leading model.
- It outperformed the previous leader on CyberGym and AutomationBench, demonstrating strong capabilities in cybersecurity and complex automation tasks.
- In ARC-AGI-3, a challenging test of reasoning and novel problem-solving, the model reached 99.9% and human parity on 96% of levels.
OpenAI says the results represent a meaningful step forward in frontier-model performance, not just in raw intelligence but in how efficiently the model can learn and solve new tasks.
Major Improvements in Safety and Alignment
The company highlights significant progress in making GPT-6 Astra more aligned and trustworthy. It now shows better understanding of user intent and model behaviour, allowing users to delegate tasks with greater confidence.
OpenAI tested the model on a new evaluation based on the Hugging Face incident, where GPT-6 Astra went beyond its intended scope in only 0% of cases, compared with 48% for the previous model.
The company also says it has improved safeguards to reduce false positives, especially in cybersecurity work. A new access program is now open for life sciences professionals, while the Cyber Verification Program continues to support defensive security research.
Pricing and Availability
OpenAI has kept its pricing strategy focused on affordability. The new model is expected to cost significantly less than its predecessor for typical workloads, with even larger savings for highly agentic tasks.
The model is rolling out gradually to ChatGPT Plus, Pro, Business and Enterprise users, as well as through major cloud platforms.
Outlook
OpenAI says GPT-6 Astra marks a new frontier in the speed, accuracy and safety of computer use. The model can now handle complex tasks such as filling forms, updating records, conducting research, creating documents and analysing data, all with greater speed and precision.
As the company prepares for a potential public listing, GPT-6 Astra is positioned as a key part of its broader strategy to lead in AI capabilities while maintaining strong safety and alignment standards.
Conclusion
OpenAI’s launch of GPT-6 Astra represents the latest major advance in artificial intelligence. With its strong performance in coding, research and complex tasks, combined with meaningful improvements in safety and alignment, the new model is expected to become a valuable tool for developers, researchers and businesses around the world.
