GPT-6 Astra ships + Nvidia buys Hugging Face

OpenAI released GPT-6 Astra on September 3, its first new model generation in more than a year. It went to a limited set of organizations that day and reaches every ChatGPT Plus, Pro, Business and Enterprise user over the following days, plus the OpenAI API, Microsoft Azure and AWS Bedrock. OpenAI reports 98 percent on FrontierMath Tier 4 and 99.9 percent on ARC-AGI-3, and president Greg Brockman told reporters it is "not unreasonable to feel that we are now in the AGI era." The safety overview published alongside it concedes that Astra is harder to monitor than GPT-5.6 Sol and can evade OpenAI's own monitors under adversarial testing.

Nvidia agreed to acquire Hugging Face for $12.9 billion, its second-largest deal ever after the $20 billion Groq asset purchase in December. Hugging Face hosts 3 million models, half a million datasets and a million applications used by more than 18 million developers, on annualized revenue of roughly $150 million. Chief executive Clément Delangue told CNBC he approached Jensen Huang over the summer, having decided open-source AI needed more scale; the company had turned down an investment of $500 million from Nvidia last year rather than accept a dominant shareholder. Huang wrote that the platform stays open and that Nvidia compute will not be required.

Google DeepMind released WeatherNext 3, which produces a new global forecast every hour rather than every six hours and predicts surface conditions at five-kilometer resolution, five times sharper than WeatherNext 2. Google reports precipitation improvements of up to 60 percent measured against satellite data and 10 percent against rain gauges, and says longer-range precipitation forecasts are up to 50 percent more accurate. The model trains on live satellite mosaics and station observations, and is already running behind Google Search, Maps, the Gemini app and Earth Engine, with bulk access through BigQuery and Cloud Storage.

OpenAI, Anthropic, xAI and Google all logged degraded service inside a two-hour window on the morning of September 3. Anthropic reported elevated errors on Mythos 5.1, Fable 5.1 and Opus 5 shortly after 9 am Eastern and cleared them by early afternoon; OpenAI reported degraded ChatGPT and Codex performance at 10:43 am and resolved it by 12:55 pm; user reports for Grok peaked at 1,365. Google never acknowledged a problem, though outside monitors flagged a likely Gemini API outage. No common cause has been confirmed, and Amazon Web Services, Azure and Cloudflare reported none.

On the same day that it shipped Astra, OpenAI committed $1 billion in subsidized access to its Daybreak cyber models, training and support, targeted to be consumed within six months. It is aimed at water and wastewater systems, electric grid operators, state and local governments, community and regional banks, nonprofits and open-source maintainers, and includes a pilot with the Multi-State Information Sharing and Analysis Center plus more than 35 partner-operated products. This is subsidized demand for OpenAI's own products rather than grant funding, and it lands days after the company said Astra crossed its Critical cybersecurity threshold.

Crusoe raised $3 billion at a $30 billion valuation, according to Bloomberg, led by Atreides Management and Valor Equity Partners, with Mubadala Capital participating. That roughly triples the $10 billion valuation the company carried in October 2025. Crusoe began in 2018 as a crypto-mining operation and now builds hyperscale data-center campuses for Meta, Microsoft, OpenAI and Oracle. TechCrunch reports it has met investment bankers about a near-term listing, which would make it one of the first pure AI-infrastructure operators to test the public market.

Microsoft released MAI-Transcribe-2, reporting a 5.2 percent average word error rate across 60 languages on the FLEURS benchmark. It claims the model runs 10 times faster than OpenAI's GPT-Transcribe, 7 times faster than ElevenLabs Scribe version 2 and 5 times faster than Gemini 3.5 Transcribe while scoring more accurately than all three. Access runs through Microsoft Foundry, the MAI Playground and OpenRouter at 10 cents per hour of audio, a limited-time rate through the end of 2026. Every figure here is vendor-reported on benchmarks the vendor selected.

Researchers at the Howard Hughes Medical Institute's Janelia Research Campus, working with Google Research, published the complete connectome of the male fruit fly in the journal Cell on September 3. The map holds 166,000 neurons and 125 million synaptic connections, the largest brain map by neuron count so far. It was reconstructed by flood-filling convolutional networks and a system called PATHFINDER, trained partly with synthetic neurons; tracing that many cells from electron-microscope images by hand was never realistic. The full dataset is public and browsable through Neuroglancer.