- Published on
Just Now, GPT-6 Shockingly Released! Humanity Enters the Great Division of Labor in the AGI Era
- Authors
- Name
- DeepHorus
OpenAI has officially released its new generation AI model, GPT-6 Astra, which the company has dubbed "the world's most intelligent and human-intention-aligned model." This release marks the largest model upgrade by OpenAI in recent years, with OpenAI President Greg Brockman even stating that looking back, today may be the starting point of the AGI (Artificial General Intelligence) era.
The core breakthrough of GPT-6 Astra lies in its powerful "Computer Use" capability. Unlike previous models focused on chat and Q&A, Astra is more like an AI agent capable of directly executing tasksâit can call tools, operate software, browse the web, and independently complete complex work. According to OpenAI, it can fill out online forms, update CRM data, operate Excel and Power BI, conduct web research, and even use KiCad to design PCB circuit boards, create 3D models with Blender, and import them into Unreal Engine 5 to generate interactive scenes.
In benchmark tests, Astra demonstrated significant performance improvements. In the OSWorld 2.0 computer operation test, Astra scored 72.6%, surpassing the previous model GPT-5.6 Sol's 65.7%. In the Agents' Last Exam test, which assesses agentic capabilities, Astra scored 59.3%, also leading over Claude Opus 5 and GPT-5.6 Sol. Its software engineering capabilities were also recognized, with a score of 57.7% in the Terminal-Bench 4.0 test, far above GPT-5.6 Sol's 37.3%.
The release of Astra aims to integrate AI deeply into core enterprise workflows beyond just being an auxiliary tool. In scientific research, Astra achieved an excellent score of 97.6% in the FrontierMath Tier 4 test and directly participated in mathematical research, helping advance the problem of prime gaps. Partner Jane Street noted that code generated by Astra in agent programming scenarios is easier to understand and requires fewer iterations to reach production quality.
However, Astra's exceptional capabilities have also sparked significant safety concerns. In the ExploitBench cybersecurity test, Astra achieved a perfect score of 100% and was found to have discovered and exploited two previously unknown zero-day vulnerabilities during the test. OpenAI acknowledged that Astra has reached the "critical cybersecurity capability threshold" in its safety framework, so it has adopted a stricter tiered approach for public release, with the general user version refusing certain advanced security requests.
Regarding commercialization, the API pricing for Astra is set at $10 per million input tokens and $50 per million output tokens, which is a notable increase from previous models. OpenAI believes that future industry metrics will shift from "token pricing" to "task cost." Astra will first be available to enterprise users in the Daybreak program, then gradually rolled out to paying users on ChatGPT Plus, Pro, and other tiers.
Greg Brockman views the release of Astra as a historical milestone, describing it more as a "mission concept." Although OpenAI has not officially announced the achievement of AGI, Astra's demonstrated ability to autonomously understand goals, plan pathways, and proactively complete work marks AI's evolution toward higher levels of autonomous forms. In the future, the mode of task division between humans and AI may undergo a fundamental transformation.
