According to OpenAI president Greg Brockman, the model rolls out initially to a limited number of organizations before expanding to paid ChatGPT subscribers in the coming days, while free users maintain no access.
Cybersecurity Risks and Maximum Threat Classification
OpenAI classifies Astra under its highest security risk tier following a summer marked by spontaneous hacking incidents involving AI tools from major developers, according to company statements. To mitigate these threats, Astra deploys automated surveillance mechanisms capable of interrupting unauthorized tasks in real time. The company restricted the model’s most sensitive cyberdefense functions to specialized security professionals. In parallel, rival firm Anthropic distributes its powerful Mythos model to select partners while offering a restricted version named Fable to the general public, as reported by industry disclosures.
Regulatory Landscapes and Federal Oversight
United States federal law currently places no direct legislative restrictions on advanced AI models, though a voluntary framework exists via an executive order signed in June by President Donald Trump, which established federal safety examinations. According to Greg Brockman, GPT-6 completed this voluntary federal review with no flags raised. International discussions at the G20 summit in North Carolina revealed a stark divergence in policy, with Washington pushing back against new regulatory oversight bodies while OpenAI chief scientist Jakub Pachocki advocated for shared safety standards and international coordination.
Autonomous Computing and Pricing Models
Beyond security evaluations, OpenAI positions GPT-6 as an autonomous computing agent capable of executing complex multi-step workflows. Company testing shows the model can independently locate housing options in ten minutes—a task requiring six hours for human testers—and has achieved two independent mathematics discoveries. Within developer environments like Codex, Astra resolves prompts with less rigid instruction steps while maintaining lower execution costs per task than its predecessor, according to release documentation. OpenAI lists API access for the model at $10 per million input tokens and $50 for output tokens, matching rival pricing structures set by Anthropic.
Worth a look