SYS // WINDYVIEW
NEURAL FEED
TRANSMISSION
VERIFIED
--:--:--
AI DIGEST
2026-09-07
← BACK TO ARCHIVE

// NODE-02 · NEURAL FEED · DAILY TRANSMISSION //

AI NEWS
DIGEST

// TOP STORIES //

1. OpenAI Ships GPT-6 Astra, Claims the AGI Era Has Started

OpenAI released GPT-6 Astra on September 3, calling it "the world's most intelligent and aligned model." It was trained on more than 100,000 GPUs at the Stargate site in Texas — OpenAI's largest run ever, and the first where earlier models supervised the training of the new one. Astra posts 97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3 and 72.6% on OSWorld 2.0 against 65.7% for GPT-5.6 Sol, with a 1M-token context window at $10/$50 per million input/output tokens. During cybersecurity evals it found and used two previously unknown zero-days, which OpenAI is disclosing to maintainers.

Source: Vellum

2. Claude Fable 5.1 and Gemini 3.8 Flash Land in the Same Week

The frontier labs answered within 24 hours of each other. Anthropic shipped Claude Fable 5.1 on September 3 at unchanged $10/$50 pricing but with cache reads cut to $0.25 per million tokens, and a self-reported Terminal-Bench-Science score of 52.6. Google followed on September 4 with Gemini 3.8 Flash and a Flash Cyber variant, plus the Lyria 3.5 music model in the Gemini app and API. Meta also pushed Muse Spark 1.3 the same week.

Source: LLM Stats

3. Claude Produces the First Computer-Checked Proof of Fermat's Last Theorem

Anthropic says Claude formalized Fermat's Last Theorem in Lean over 11 days, working largely autonomously — a task mathematicians had expected to take several years. The run generated 13 million lines of Lean code, proved 30,300 theorems (29,500 used in the final proof) and burned roughly six billion output tokens. The breakthrough came after Claude was given an open-source tool called Prove2Me. It did not reprove Wiles's 1995 result; it converted it into a form a machine can verify end to end.

Source: Anthropic

4. Anthropic Locks In $517B of Compute Ahead of a Possible IPO

Anthropic has signed agreements over the past 11 months securing at least 14.8 gigawatts of computing capacity, with total spending over the next decade potentially reaching $517 billion across cloud leases, chip procurement and data centre leases. Google and AWS account for about 11 GW between them, including a single 11-gigawatt arrangement worth more than $300 billion over ten years. The company is committing to capacity years ahead of the revenue needed to pay for it.

5. EU Begins First AI Act Inspections; California SB 1047 Hits Newsom's Desk

The EU AI Act became applicable on August 2, and this month the European AI Office in Brussels — working with 24 national market surveillance authorities — starts its first scheduled wave of compliance inspections. By September 15, providers of general-purpose models trained above the 10^25 FLOPs threshold must file their first formal systemic-risk evaluations. In the US, California's SB 1047 cleared both chambers in late August and Governor Newsom has until midnight on September 30 to sign or veto it, while federal preemption stays stalled in the House Judiciary Committee.

Source: Cubbbix

6. US Probes $3B in Nvidia Blackwell Chips Routed Through Southeast Asia

Aivres, the US-based subsidiary of Chinese server maker Inspur, is under investigation over more than $3 billion in Nvidia Blackwell chips exported to Southeast Asia between April 2024 and February 2026. The case is the largest export-control question yet raised about the Blackwell generation and lands while Washington is still arguing over how tightly to police downstream resale of advanced accelerators.

Source: AI Weekly

7. Google Starts Switching Off Google Assistant for Good

On September 4 Google began removing access to Google Assistant across Android phones, tablets, Wear OS watches and Android Auto, handing 'Hey Google' triggers and the power-button long-press to Gemini. The rollout is gradual but one-way and irreversible — the clearest signal yet that Google is done maintaining two assistant stacks.

Source: AI Weekly

8. Uber Puts $100M Into Travis Kalanick's Robotaxi Startup

Atoms, the robotaxi venture from Uber founder Travis Kalanick, raised a $1.7 billion round led by Andreessen Horowitz, with Uber itself investing $100 million. The deal puts Uber money behind its own founder's autonomy play at a moment when robotaxi economics are finally being tested at city scale rather than in pilot geofences.

Source: AI Weekly

9. Huawei's Mate XT2 Claims a US-Free Supply Chain

Huawei launched the triple-folding Mate XT2 on September 7 at RMB 19,999, built on the Kirin 9050 Pro and claimed by the company to be "entirely free of US supply-chain restrictions," with a stated 42% performance gain over the prior generation. Whether the claim holds up matters well beyond phones — it is a proxy for how far China's domestic silicon has actually closed the gap for on-device AI.

Source: AI Weekly

// KEY TAKEAWAYS

Three frontier models landed in 48 hours — GPT-6 Astra, Claude Fable 5.1 and Gemini 3.8 Flash — and the interesting number is no longer the benchmark score but the bill behind it: Anthropic alone has committed up to $517 billion in compute against revenue it has not yet earned. Capability is showing up in places that are hard to hand-wave, like an 11-day machine-checked proof of Fermat's Last Theorem and zero-days discovered during a safety eval. Meanwhile the regulatory clock finally started ticking for real: EU inspectors are in the field this month, GPAI systemic-risk filings are due September 15, and California's SB 1047 sits on Newsom's desk until September 30.