GPT-6 Astra launches, under $6 an hour AI engineer

GPT-6 Astra represents a shift from models that assist engineers to ones that can act as engineers themselves. In early access, the authors—who admit they aren't qualified to assess scientific benchmarks—focused on practical tasks, burning over 20B tokens. They found Astra could handle a wide range of real-world projects, from building internal tools to replacing paid SaaS services and training game AI for a board game with 10,000x more legal moves than Go. This capability, combined with an effective cost under $6 per hour, challenges the economics of hiring junior engineers.

The model's performance on benchmarks like FrontierMath (97.6%) and ARC-AGI-3 (99.9%)—completely saturating the hardest versions—underscores its reasoning power. OpenAI's own leaders, Greg and Jakub, have made strong claims: Greg says AGI is here, and Jakub describes it as the Automated AI Research Intern he wanted. These endorsements, while subjective, suggest internal confidence.

For businesses, the implication is not that all engineering jobs vanish overnight, but that the cost of AI-driven development could drop significantly. The authors note they spent about $100 over two days to accomplish what they'd previously pay a junior engineer for. This points to a near-term future where leveraging models like Astra is not optional but essential.