OpenAI’s GPT-6 Astra: What the reported “AGI era” claim means for users, developers, and security teams

OpenAI has shipped GPT-6 Astra and describes it as its most capable model to date. Greg Brockman said the release may already qualify as “AGI” or at least put that threshold within reach, using OpenAI’s definition of a system that outperforms humans at most economically valuable work. The model is rolling out first to select organizations through OpenAI’s Daybreak program, with broader access expected for ChatGPT Plus, Pro, Business, and Enterprise customers in the coming days. API access and availability through cloud platforms including AWS Bedrock and Microsoft Azure are also planned.

OpenAI’s published benchmarks put Astra ahead of GPT-5.6 Sol and Anthropic’s Fable models across several key areas. Reported highlights include 97.6 percent on FrontierMath Tier 4 v2, 74.1 percent on DeepSWE v1.1, 96 percent on GPQA Diamond, 95.9 percent on BenchCAD, and 100 percent on ExploitBench. The model also scored 72.6 percent on OSWorld 2.0, compared with Sol’s 65.7 percent, while taking about 40 minutes per task versus roughly 75 minutes for Sol. For readers evaluating AI tools, the takeaway is that Astra is being framed less as a chatbot upgrade and more as an agentic work system for software, research, engineering, and computer operation.

GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens in standard API mode. A fast mode promises 2.5x speed but doubles the price, putting Astra at 2.5x the cost of GPT-5.6 Sol and in the same range as Anthropic’s Fable 5.1. Brockman argued that token prices are becoming a weak comparison metric because tokenization differs between models and providers. OpenAI’s preferred framing is cost per completed task, and the company says Astra’s top DeepSWE v1.1 configuration cuts estimated API costs per task by about 57 percent versus Sol.

Astra is the first model OpenAI classifies as “critical” under its Preparedness Framework. That classification means the model can find previously unknown vulnerabilities and build exploit chains across well-defended systems when given the right tools and access, without step-by-step human guidance. During evaluation, Astra reportedly found two previously unknown zero-day vulnerabilities, which OpenAI reported to affected vendors. The strongest cybersecurity capabilities are currently restricted to trusted defenders in the Daybreak Blue program, underscoring the dual-use tension: the same abilities can help defenders patch systems or help attackers exploit them.