OpenAI launches GPT-6 Astra, the AI model built to do more than answer questions

OpenAI has unveiled its latest flagship AI model, GPT-6 Astra, on September 3, positioning it as the world's most 'intelligent and aligned' model. The advanced system combines years of research and significant investments across pre-training, reinforcement learning, and alignment, setting new standards in computer use, browsing, software engineering, cybersecurity, science, and professional work.

Unlike previous chatbots, Astra is designed to execute complex multi-step tasks involving browsing websites, working with documents and software, writing code, and conducting research. It can interact with computers much like a human user, filling out forms, updating CRM records, organizing calendars, conducting online research, creating documents, and even testing software.

With a 1.05-million-token context window and the ability to generate up to 128,000 output tokens, GPT-6 Astra outperforms previous models in benchmarks, achieving 98% in advanced mathematical problem-solving, 99.9% on unfamiliar reasoning problems, and 100% on cybersecurity assessments. OpenAI's President Greg Brockman hinted at GPT-6 Astra's potential as a milestone in achieving artificial general intelligence (AGI), describing it as a "mission concept or spiritual concept."

While the model promises significant benefits for users, such as delegating tasks in real-time, it also raises concerns about potential risks, including enhancing cybersecurity capabilities that could be misused by attackers and the need for ongoing safety evaluations.