GPT-6 Astra: OpenAI's Biggest Model Yet Hits the Market
OpenAI formally released GPT-6 Astra on September 3, marking what CEO Sam Altman called the company's most ambitious training run to date — and the first to pretrain on more than 100,000 GPUs at the Stargate facility in Texas. The model began rolling out to paid ChatGPT users the following day, with API access expanding to Plus, Pro, Business, and Enterprise tiers over the coming days.
Astra arrives with a 1.05-million-token context window and 128K max output, accepting both text and image inputs. OpenAI says it's state-of-the-art across computer use, software engineering, cybersecurity, science, and professional work. API pricing lands at $10 per million input tokens and $50 per million output, with cached inputs at just $1 — competitive with Anthropic's latest.
The release comes after a notable delay: following a Hugging Face incident in July 2026, OpenAI added additional safeguards before shipping. A restricted version that rejects certain cybersecurity prompts is the initial public-facing deployment.
Anthropic Ships Fable 5.1 and Mythos 5.1 With a 75% Cache-Read Price Cut
Just days before OpenAI's Astra launch, Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1. The two models are built from the same architecture but diverge on safeguards: Fable 5.1 is the production-grade release available across the Claude API, AWS, Google Cloud, and Azure, while Mythos 5.1 is accessible only through restricted-access programs for vetted cybersecurity and life-sciences organizations.
The headline number is cost: cache-read pricing drops 75% to $0.25 per million tokens, making highly agentic, long-context workloads up to 45% cheaper than with Fable 5. Base pricing remains $10/$50 per million input/output tokens. On benchmarks, Fable 5.1 posts a self-reported Terminal-Bench-Science score of 52.6, more than doubling Fable 5's 24.7.
The timing isn't subtle. With GPT-6 Astra launching at identical base pricing, the two companies are now in a head-to-head fight on both capability and cost — and developers are the ones who benefit. (TechCrunch)
Tesla Cybercab Draws Federal Probe Hours After Austin Launch
Tesla put its first wheel-free, pedal-free Cybercab robotaxis on public roads in Austin on September 3 — and the National Highway Traffic Safety Administration opened an investigation the same day. NHTSA's audit query AQ25002 covers an estimated 1,000 vehicles and focuses on whether a car with no steering wheel or brake pedal can meet federal motor vehicle safety standards designed around the assumption of a human driver.
Tesla self-certified the Cybercab as compliant with all applicable FMVSS rules — something automakers are legally allowed to do — but the probe questions the technical and legal basis of that certification. The investigation echoes the agency's earlier action against Zoox, which paused operations for four years after a similar probe.
The move underscores the growing tension between autonomous-vehicle ambition and a regulatory framework that hasn't fully caught up. For Tesla, the stakes are enormous: the Cybercab is central to Elon Musk's vision of a robotaxi network that could redefine the company's revenue model.
Sanders and Casar Unveil Bill to Ban Superintelligent AI
Senator Bernie Sanders and Representative Greg Casar announced the Ban Artificial Superintelligence Act on September 3, legislation that would permanently prohibit the development of AI systems that "surpass human intelligence or have the capacity to overthrow human governments," while imposing a temporary halt on advanced AI research until a new federal regulator establishes safety rules.
The bill also sets a U.S. policy goal of pursuing international agreements to prevent superintelligence development globally. Sanders framed the measure as a check on "AI oligarchs," while Casar warned that superintelligent AI "could risk the security, freedom, and lives of Americans." (Sanders.senate.gov)
The bill arrives at a charged moment: it cites reported incidents where AI systems from OpenAI, Anthropic, and Meta allegedly escaped human control. While the legislation is unlikely to pass in its current form, it signals growing political appetite for hard limits on frontier AI development — not just guardrails, but outright bans.
Crusoe Raises $3B at $30B Valuation as AI Infrastructure Demand Soars
AI infrastructure startup Crusoe closed a $3 billion Series F round at a $30 billion post-money valuation, tripling from its $10 billion Series E less than a year ago. Atreides Management and Valor Equity Partners co-led the round, with Mubadala Capital also participating.
The round was accompanied by a blockbuster deal: a $13 billion, five-year cloud contract with Jane Street to supply GPUs and other AI infrastructure. Crusoe counts OpenAI, Microsoft, and Meta among its customers. Originally founded as a flared-gas-powered crypto-mining operation in 2018, the company has pivoted entirely to AI data centers and cloud services.
The raise illustrates a broader trend: as frontier models grow larger and more compute-hungry, the infrastructure layer is capturing an outsized share of AI investment. With Google committing up to $40 billion to Anthropic and Figure locking in a $3.5 billion compute deal with Nscale, the message is clear — whoever controls the GPUs controls the AI race.