OpenAI launches GPT-6 Astra and says welcome to the "AGI era"
Key Points:
- OpenAI launched GPT-6 Astra, its most advanced and aligned AI model to date, with OpenAI President Greg Brockman suggesting it marks the beginning of the AGI (Artificial General Intelligence) era, though the company stops short of formally declaring AGI achievement.
- Astra was trained using OpenAI’s largest training run, involving over 100,000 GPUs, and incorporates supervision from earlier models; initial rollout targets enterprise customers with broader availability to Plus, Pro, Business, Enterprise users, and API clients forthcoming.
- Pricing for Astra’s API is set at $10 per million input tokens and $50 per million output tokens, higher than some competitors, but OpenAI emphasizes task efficiency and reduced token use may offset costs.
- Astra shows significant performance improvements in coding, scientific research tasks, and cybersecurity, including high benchmark scores and the ability to discover software exploits, but access to advanced cybersecurity capabilities is restricted to vetted users via the Daybreak Blue program.
- Despite improved alignment, Astra’s reasoning is harder to monitor, prompting OpenAI to withhold further scaling until monitoring confidence improves; users may experience slowdowns or blocks, especially during cybersecurity-related tasks, at launch.