OpenAI Releases GPT-6Astra: The Industry Enters the AGI Era, with Performance Advancements Across Multiple Dimensions and a Reimagined Security Boundary

On September 3, OpenAI officially launched its latest cutting-edge large model, GPT-6Astra, which it characterizes as a "generation leap" in fields such as software engineering, professional work, science, and cybersecurity. Greg Brockman, President of OpenAI, clearly stated at the press conference that the release of Astra marks the industry's entry into the "General Artificial Intelligence (AGI) era."

OpenAI

Previously, the internal version of this model successfully solved ten long-standing problems in the fields of mathematics and theoretical computer science, and made significant breakthroughs in agent coding and cybersecurity capabilities, becoming the first OpenAI model to reach the "critical" cybersecurity capability threshold in its "preparedness framework."

In terms of technical performance and benchmark testing, Astra achieved a disruptive improvement compared to previous generations and similar models. In the ARC-AGI-3 test, Astra scored an impressive 98.6%, far exceeding GPT-5.6Sol (7.8%) and Claude Opus5 (30.2%); it scored 97.6% in FrontierMath Tier4(v2), significantly leading other large models; and it led in several tests including GPQA Diamond (96%), DeepSWE v1.1 (74.1%), Terminal-Bench Science0.1 (64.6%), Agents' Last Exam (59.3%), and AutomationBench (41.4%).

Additionally, the security protection system has been significantly enhanced. The model not only comes with new monitoring functions to prevent unauthorized operations but can also more reliably reject harmful network requests.

Currently, GPT-6Astra is available to enterprise cybersecurity customers who have gained access through the OpenAI Daybreak program, with its core security features still limited to trusted defenders. Over the next few days, it will be gradually released to ChatGPT Plus, Pro, Business, and Enterprise subscribers. In terms of pricing, Astra is set at $10 per million input tokens and $50 per million output tokens, matching the price of Anthropic's Claude Fable5.1. As an important milestone in commercialization and technological evolution, Astra's commercial deployment not only marks an increase in the threshold for advanced AI agent deployment but also opens up a new landscape of competition for large models with high computing power and high security barriers.

Altman Confirms OpenAI's In-House Humanoid Robot: Home Use Is Not the Top Priority, But Long-Term Prospects Are Promising

Sam Altman, CEO of OpenAI, confirmed that the company is advancing robot development, with parallel efforts in humanoid robots and specialized robots for data centers. He emphasized that form is not the key factor, but rather the 'robotic brain'; the company will develop humanoid robots, but if the humanoid structure lacks practical utility, it will focus on non-humanoid robots tailored for specific scenarios.

OpenAI Invests $1 Billion to Build AI Cybersecurity Barriers for Global Critical Infrastructure

OpenAI launches 'Daybreak for Frontline Defenders,' investing $1 billion over six months to provide Daybreak access, training, and support to under-resourced institutions, helping them use AI against cyber threats, prioritizing water, power, local government, and finance; includes 'Daybreak for America.'....

Global Top Three AI Platforms Suffer Mass Outage: ChatGPT, Claude, and Grok Suddenly Experience Large-Scale Downtime, All Now Fully Recovered

On September 3rd, the global AI industry experienced a rare collective failure. Mainstream large models such as OpenAI's ChatGPT, Grok, and Claude suffered large-scale outages almost simultaneously, affecting many users' daily usage and workflows. The incident impacted leading large model platforms, drawing significant attention from the industry and users. Technical teams have urgently investigated and repaired the issue.

OpenAI officially releases GPT-6 Astra: A milestone in the pursuit of artificial general intelligence

OpenAI launches the flagship large model GPT-6 Astra. CEO Brockman states that it may mark humanity's entry into the era of artificial general intelligence (AGI). The model achieves a leap in mathematical and scientific tests, with FrontierMath Tier4 and GPQA Diamond reaching 97.6% and 96% respectively. Its hands-on capabilities also perform outstandingly.

Ultraman Officially Announced: OpenAI Will Start Developing Humanoid Robots in-house, with the First AI Hardware and IPO Process Revealed

Ultraman, CEO of OpenAI, made his first clear statement that the company will firmly develop humanoid robots in-house, while simultaneously working on various forms of robots and special robots for data centers. He believes that the human physical space is adapted to the human body structure, making humanoid robots an important direction, indicating that OpenAI is accelerating its entry into the physical world.