OpenAI has unveiled GPT-6 Astra, its latest artificial intelligence model, beginning its rollout immediately following the previous generation’s introduction a year prior. The company describes Astra as exceptionally intelligent and well-aligned with human values. However, this advancement comes with significant responsibilities, as the new system represents a departure in risk profile from earlier versions.
Days before the official announcement, OpenAI disclosed that Astra had received a critical cybersecurity designation under its internal risk assessment framework, marking the first instance of any OpenAI model reaching this highest threat level. The company proceeded with the launch while implementing protective measures designed to prevent both unauthorized exploitation of the model’s capabilities and unintended harmful actions. Advanced cybersecurity functions will remain restricted to approved testing partners only, with enhanced security protocols in place across the system.
Despite these concerns, OpenAI maintains that Astra demonstrates improved safety characteristics compared to previous iterations in several key areas. The model shows reduced rates of factual inaccuracies and performs better on evaluations measuring emotional manipulation risks and age-appropriate interactions. However, the company acknowledged decreased transparency in monitoring the model’s reasoning processes compared to its predecessor. Access will roll out gradually to ChatGPT Plus subscribers and enterprise clients via API integration with pricing set at ten dollars per million input tokens and fifty dollars per million output tokens.
Benchmarking results indicate substantial improvements across most standard performance metrics, including nearly perfect scores on advanced mathematics and general knowledge assessments, though the model scored lower on certain specialized evaluations.
Originally reported by Mashable →
About the Author
