Published by Dogpay ·
GPT-6 Astra Sparks AGI Debate as OpenAI’s Safety Warnings Collide with AI Expansion
Trend One: GPT-6 Astra Triggers AGI Debate — OpenAI's Internal Contradiction Surfaces
On September 4, OpenAI released GPT-6 Astra, its most powerful model to date. NVIDIA CEO Jensen Huang posted on X that "AGI has arrived," and OpenAI President Greg Brockman declared in a conference call, "Welcome to the AGI era." Astra was reportedly trained on over 100,000 NVIDIA Grace Blackwell NVLink72 GPUs, supports autonomous control of professional software such as Blender, Photoshop, and Unity, and has reduced hallucination rates to approximately 2%. Huang revealed that the next batch of 400,000 GPUs is about to come online.
On September 6, CNBC reported on growing "model fatigue" in the industry. Within a single week: Anthropic updated Fable and Mythos (September 1), Meta released Muse Spark 1.3 (September 2), Google launched Gemini 3.8 Flash (September 2), OpenAI released GPT-6 Astra (September 4), and Abu Dhabi's MBZUAI open-sourced its K2 Horizon series. Gartner projects 2026 AI spending at $2.59 trillion, up 47% from 2025, with over half going to infrastructure. Runpod CEO Zhen Lu noted that model fatigue "has definitely become a real problem," while Clockwork Systems CEO Suresh Vasudevan observed that "every release now is so insanely strong that it's become hard to judge what counts as a true leap forward."
Meanwhile, Anthropic's Claude completed the first end-to-end formalized proof of Fermat's Last Theorem in the Lean proof assistant, running autonomously for 11 days. The system generated complete Lean code from the proposition statement through kernel type-checking, without human intervention on proof decomposition. Imperial College mathematician Kevin Buzzard commented that this represents "a major step toward automatically formalizing the entire modern mathematical literature."