The first week of September delivered one of the most consequential stretches in recent AI history, with OpenAI and Anthropic effectively trading blows in real time while Wall Street quietly priced in a trillion-dollar valuation for one of them.

GPT-6 Astra crosses a threshold OpenAI once called a red line

On September 3, OpenAI introduced GPT-6 Astra, its first full-generation model since GPT-5. OpenAI described it as the most capable model it has ever broadly deployed, and the first to reach the Critical level of cybersecurity capability under its Preparedness Framework. That's not a marketing flourish — it's an internal safety classification, and OpenAI itself acknowledged the stakes: with the right tools and access, the model can find previously unknown security flaws and develop new ways to exploit them across many well-protected systems without a person guiding each step.

The capability jump wasn't limited to security. OpenAI said Astra was built on its largest-ever training run, using more than 100,000 GPUs at its Stargate site in Texas, and that this is its first model to use other models in a significant role in supervising Astra's training. On the science side, the model helped improve a mathematical result on gaps between prime numbers and set new marks on several biology, chemistry, medical and physics evaluations. OpenAI president Greg Brockman went further in the press briefing, closing with the line, "Welcome to the AGI era." Nvidia's Jensen Huang echoed the sentiment days later, though plenty of researchers — Gary Marcus among them — pushed back, noting there's no agreed-upon definition of AGI in the first place.

Anthropic answers with Fable and Mythos — and a compute spending spree

Just two days before Astra shipped, Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1, models it says represent the strongest overall cyber capabilities of any model it has released. The two are functionally identical except for guardrails: Mythos is Fable with more permissive safeguards for vetted individuals and organizations affected by cybersecurity and life-sciences restrictions. Fable itself got materially cheaper — an estimated 25% cost reduction for typical workloads, and up to roughly 45% savings for highly agentic work — while cybersecurity false positives reportedly dropped sharply, easing a common enterprise complaint.

That launch came on the back of eye-popping infrastructure spending. Days earlier, Anthropic signed a roughly $45 billion deal to rent about 460 megawatts of computing capacity in a West Virginia data center equipped with Nvidia's latest Rubin GPUs and Vera CPUs. That's on top of a separate $35 billion infrastructure deal with cloud startup Lambda.

Why it all matters: an IPO is coming

The spending spree isn't happening in a vacuum. Anthropic filed a draft registration statement with the SEC on June 1, with its most recent private valuation at roughly $965 billion, and press reports floating a $1 trillion-plus base case if the company lists this fall. More recent reporting has pushed that number even higher — the New York Times reported in late August that an eventual IPO could value the company at $2 trillion, which would make it the largest ever. That would place Anthropic's debut above SpaceX's, which already looked expensive when it went public at around $1.75 trillion.

Taken together, the week captures where the industry actually stands: capability races measured in cybersecurity thresholds rather than chatbot demos, infrastructure bills running into the tens of billions per deal, and public markets about to get their first real referendum on whether any of it is worth the price tag.

Sources

AI IPOAnthropicArtificial IntelligenceDaily AI UpdateGPT 6OpenAI