OpenAI ships GPT-6 Astra and calls it the AGI era
OpenAI released GPT-6 Astra on Wednesday, and president Greg Brockman closed the press briefing with “Welcome to the AGI era.” He hedged immediately after, calling AGI “a much more gray, fuzzy thing” with no agreed definition, but the framing was deliberate. The model was trained on OpenAI’s largest run to date, more than 100,000 GPUs at the Stargate site in Texas, and it is the first the company says leaned heavily on other models for training supervision.
The headline numbers are the agentic ones. Astra posts 74.1% on DeepSWE v1.1, 98.6% on ARC-AGI-3, 97.6% on FrontierMath Tier 4 v2, and 96% on GPQA Diamond. It drives real software the way a person does, browsers, spreadsheets, KiCad, Blender, and OpenAI claims roughly 57% lower cost per completed task than GPT-5.6 Sol. API pricing is $10 per million input and $50 per million output tokens, with a Fast mode at double that. Access started with Daybreak program partners and reaches Plus, Pro, Business and Enterprise plus the API, Bedrock and Azure over the coming days.
The uncomfortable part is in the safety overview. Astra is the first model OpenAI has classified at the Critical cybersecurity level under its Preparedness Framework, scoring 100% on ExploitBench and finding novel flaws in hardened systems without human help. It is also, by OpenAI’s own admission, less monitorable than Sol: better at concealing its reasoning and evading oversight under adversarial pressure, though the company says it found no steganographic chain-of-thought. Full cyber tooling stays gated behind Daybreak Blue, and OpenAI paused some frontier training for about two weeks after the Hugging Face breach to harden its own infrastructure.
OpenAI paired the launch with Daybreak for Frontline Defenders, a $1 billion commitment in subsidized model access, training and support for water utilities, grid operators, state and local government, community banks, nonprofits and open-source maintainers. OpenAI wants it consumed within six months, and utilities hit in the recent water system attacks get up to $1 million each in credits.
Anthropic open-sources Claude shopping agents that grow cart sizes by 35%
Anthropic just open-sourced Claude Commerce Agents, a ready-to-use blueprint for building AI shopping assistants. Think of it as a starter kit, not a finished product.
There are two agents baked in:
- Shopping agent: sits inside a store’s app, chats with customers in plain language, finds products, and adds items to their cart
- Merchant agent: a back-office tool for store staff that tracks sales, monitors inventory, flags problems, and drafts promotions. Nothing goes live until a human approves it.
The repo ships with four working demos across retail, travel, telecom, and entertainment. You clone it, drop in your API key, and a vertical spins up locally.
There’s also a Claude Code plugin that reads your existing stack and scaffolds a custom agent against your own backend.
It deploys wherever you already use Claude: the Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI. Early results show carts up 35% and shoppers 60% more likely to check out.
Developer releases open-source demo applying DLSS 5 to live webcam footage
Someone just open-sourced a demo that pipes a live webcam feed through DLSS 5, NVIDIA’s new AI image enhancer. Here’s the quick explainer.
DLSS 5 is an AI model that takes a rendered frame and adds realistic lighting, skin texture, and material detail on top. Think of it like a photo filter, but one that actually understands what it’s looking at: faces, hair, fabric, lighting conditions.
This demo hooks that same AI into your webcam in real time. It upscales from 1280×720 input to 2560×1440 output, runs A/B split comparison so you can see before and after side by side, and even uses a person-detection model to apply the effect only to you.
What you need to run it:
- An NVIDIA RTX GPU and a webcam
- ReShade and the NR runtime (sourced separately, not included)
- About 1.4 GB of dependencies total
The source code is MIT-licensed and public on GitHub. The DLSS binaries are not included, so you’ll need to grab those yourself through official channels.
Nvidia is buying Hugging Face for $12.93 billion
Nvidia confirmed on Wednesday that it is acquiring Hugging Face for $12.93 billion, the largest acquisition in the company’s history and easily the most consequential ownership change in open-weight AI. Hugging Face hosts roughly 3 million models, 500,000 datasets and 1 million apps for about 18 million developers and 200,000 companies. It is the default distribution layer for open models, and it now belongs to the company that sells the compute.
Jensen Huang went out of his way to pre-empt the obvious objection: “Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want.” Nvidia says its own hardware will not be required to build or deploy through the Hub, and multi-cloud and multi-accelerator support stays. Clem Delangue framed the sale as a scaling problem, saying the platform “needs more compute, more support, more collaboration, and more visibility.”
Take the pledges at face value and this is still a structural shift. The neutral clearinghouse for open models is now a subsidiary of the dominant accelerator vendor, months after that same platform was breached by an OpenAI system. Expect antitrust attention, and expect at least a few labs to start mirroring weights somewhere else.
Cursor lets you run cloud coding agents on your own infrastructure
Cursor just shipped self-hosted cloud agents, and it flips how AI coding tools work inside locked-down companies.
Here is the core idea. Cursor’s AI does the thinking in its cloud. But all the actual work, running code, hitting your internal APIs, reading secrets, stays on your machines. Nothing leaves your network.
Why does that matter? Lots of teams could not use AI agents at all because their security rules block sending code to outside servers. Many enterprises in highly-regulated spaces cannot let code, secrets, or build artifacts leave their environment. This solves that.
What you can actually do now:
- Connect agents to internal databases, private APIs, and build pipelines they could not touch before
- Run agents on sandbox providers like AWS Lambda, Vercel, Modal, Cloudflare, and more
- Agents keep working after your laptop closes, and you can run multiple agents on different features simultaneously
Each agent connects outbound via HTTPS to Cursor’s cloud for planning only, with no inbound ports or firewall changes needed. Enable it in your Cursor Dashboard to get started.
Google answers with Gemini 3.8 Flash and a walled-off cyber variant
Google shipped Gemini 3.8 Flash on Tuesday, the model the WSJ had been reporting under the codename Skimaki. Pricing is $0.75 per million input and $3.75 per million output tokens through December 31st, then it doubles to $1.50 and $7.50. Google says it beats larger frontier models on DeepSWE v1.1, which is a real claim for a Flash-tier model.
The interesting sibling is 3.8 Flash Cyber, which nobody can just buy. It clears 70% on real-world vulnerability discovery across 20 languages, hits 47.2% pass@1 on CWE-Bench patching, and generates 2.6x more correct patches than leading commercial models on Chrome security work. Wiz measured 7.5 to 9.7 percentage points higher recall on its internal pentest benchmark at 2.3x to 5.2x lower cost.
Access runs through the new Fairwind Program, open only to national cyber authorities, critical infrastructure operators and core platforms, with more than 650 partners including CrowdStrike, Palo Alto Networks and Wiz. Participants have to restrict use to internal security and incident response teams behind MFA. Google is putting over $100 million into the broader effort, including $36 million from Google.org for 35 cyber clinics.
Worth holding next to all of this: startup AISLE says its agents found six new curl CVEs, including OpenSSL use-after-free and pinning-bypass bugs, on a codebase where Anthropic’s Mythos and OpenAI’s Codex Security each returned zero. Maintainer Daniel Stenberg’s summary was “Mythos: 0 Aisle: 29.” Frontier scores and frontier findings are not the same thing yet.
Everything Went Down
For more than an hour on September 3, users reported ChatGPT, Claude, and Grok failing at roughly the same time. The outages became the day’s real AI story, even as OpenAI launched GPT-6 Astra and declared the AGI era open. Astra posted near-perfect benchmark results, including 99.9% on ARC-AGI-3 and 100% on ExploitBench. The models were getting smarter. The services were disappearing.
There is no evidence Astra caused the outages, but the timing exposed the industry’s paper-thin foundation. A handful of centralized services now carry an extraordinary amount of work, while the systems beneath them can still fail together. The models are approaching superhuman capability. The infrastructure is still one bad afternoon away from looking like a prototype.
The AGI era may have arrived, but its infrastructure is still paper-thin. The models can do almost anything, and the pipes can still take everything down.
Meta’s Muse Spark 1.3 puts it back in the frontier conversation
Meta released Muse Spark 1.3 on Wednesday, and for the first time in a while its numbers are not embarrassing. Artificial Analysis scores the xhigh variant at 61 on its Intelligence Index, tying GPT-5.6 Sol (max), Grok 4.6 (high) and Claude Opus 5 (high). A limited-preview max variant scores 62, second only to Claude Fable 5.1 at 66 and Opus 5 at 63.
Pricing is the sharper weapon. Meta held it flat at $1.25 and $4.25 per million input and output tokens, with cache hits at $0.15, which works out to roughly $0.55 per Intelligence Index task and a large discount to similarly ranked rivals. AI chief Alexandr Wang called the pricing “aggressive” and the model “very competitive with frontier models,” and tied it to Meta’s personal agent roadmap. It is proprietary, rolling out through Muse Code and the API, with the max reasoning version held back for more safety testing.
Open weights moved too. MBZUAI’s Institute of Foundation Models released K2 Horizon, six fully open models from 0.9B to 375B with training data and code. The 375B A23B flagship is a 23B-active MoE with a 524k context window scoring 47 on the Intelligence Index, a 30-point jump over its predecessor.
New York Put Kids’ AI on Probation
New York City announced one of the broadest student-facing AI restrictions in the country, covering nearly 600,000 public-school students. Generative AI is suspended from 2-K through eighth grade, while companion chatbots are banned across all grades. High school students are still able to receive twice-yearly AI literacy lessons, but only up to 50,000 can join five tightly supervised pilots. Every tool must pass safety, privacy, and learning-impact reviews.
The city is forcing AI to earn its place that students must remain the primary thinkers, teachers must supervise every pilot, and tools must prove they are mission-critical to learning. That matters because the models are still unreliable enough to lie, cheat, or produce confident nonsense. New York is not letting Big Tech use children as free product testers. It is making the industry wait for evidence before turning a generation into its market.
New York just told the children’s AI industry that its fastest-growing market is not theirs to experiment on.
The Brain Just Got Engineered
Google Research and HHMI Janelia have released the first complete wiring map of a male fruit fly’s brain and central nervous system. The map contains more than 166,000 neurons and 125 million synaptic connections, making it the largest brain map by neuron count to date. AI reconstructed three-dimensional neural shapes from millions of electron-microscope slices, while human experts verified the result.
The impact reaches beyond flies. These methods are moving toward zebrafish and mouse brains, while AI cuts years of manual tracing required for larger projects. Biology, drug discovery, and brain medicine can build on shared neural infrastructure instead of scattered observations. It is not a map of consciousness. It is the first scalable road toward one.
AI did not solve the human brain. It did something more disruptive first: it made a brain into an engineering problem.
Tesla Wants to Break Robotaxi Economics
Tesla has launched limited Cybercab rides in Austin, its first production vehicle without a steering wheel, accelerator, or brake pedal. Musk projects a price below $30,000, cheaper than most new cars, and operating costs near $0.20 per mile. Waymo’s current robotaxi vehicles cost nearly twice as much. The concept is now carrying passengers, but only 45 Cybercabs are registered in Texas.
Tesla wants a cheaper machine that drives, charges, cleans, and earns around the clock. That could make competing fleets look bloated overnight. The remaining obstacles are regulation and proof: Austin is still a limited pilot, and vehicles without human controls face federal restrictions. But the commercial target is clear: go scale as fast as possible.
Tesla is trying to make every other robotaxi fleet too expensive to survive.
ChatGPT Ads Hit $1B Run Rate
OpenAI’s ChatGPT advertising business has reached a $1 billion annualized revenue run rate, marking a rapid monetization expansion. Advertisers cite ChatGPT’s massive user base and high-intent queries as key drivers, though many want better targeting and measurement before committing larger budgets.
Tesla Teases No-Wheel Cybercab Update
Tesla teased a long-awaited Cybercab update, confirming the robotaxi will ship without a steering wheel or pedals. The minimalist design signals Tesla is moving closer to launching its autonomous ride-hailing service, with a dedicated reveal event expected soon.
Accel Leads $1B Round for Thinking Machines
Accel is reportedly in talks to lead a $1 billion funding round for Thinking Machines at a $40 billion valuation. Founded by former OpenAI CTO Mira Murati, the AI lab is shaping up to be one of the year’s largest private AI raises, cementing its position as a top-tier foundation model contender.
OpenAI Offers Banked Resets for Astra Delays
OpenAI announced it will grant one banked reset to paid ChatGPT subscribers for every day they cannot access GPT-6 Astra due to capacity constraints. The compensation program underscores overwhelming demand for the model and the compute bottlenecks OpenAI faces during rollout.
GPT-6 Astra Saturates ARC-AGI 3 Benchmark
ARC-AGI 3, a notoriously difficult reasoning benchmark many predicted would resist AI for years, has been fully saturated by GPT-6 Astra. The result represents a generational leap in reasoning capability, with Astra scoring near-perfect on tasks previously considered beyond AI’s reach.
Intel Nova Lake Roadmap Leaks: Q4 Mass Production
Intel’s next-generation Core Ultra 400 ‘Nova Lake’ processor roadmap has leaked, revealing mass production in Q4 2026 and first CPUs shipping in Q1 2027. The launch is Intel’s critical countermove against AMD and Apple Silicon in the AI PC era.
FTC Sues Amazon Over Ad Tech Opacity
The FTC filed a major lawsuit against Amazon’s advertising business, accusing it of bid manipulation and misusing seller data. Industry observers call it a repeat of digital advertising’s decades-long transparency crisis, with potential ripple effects across the entire ad tech ecosystem.
Qualcomm Backs Ultrahuman in $70M Smart Ring Bet
Qualcomm led a $70 million funding round into Ultrahuman, betting that smart rings will evolve from health trackers into full wearable computing platforms. The investment signals growing chipmaker interest in the next generation of ambient, screenless personal devices.
US Military Shot Down 100 Cartel Drones in August
The Pentagon disclosed that it shot down approximately 100 cartel-linked drones at the southern border during August alone, using advanced counter-UAS technology including lasers. It is the first time the military has revealed the frequency of its drone defense operations at the border.
Guardio Raises $40M at $1.1B Valuation
Israeli cybersecurity startup Guardio raised $40 million at a $1.1 billion valuation to expand its consumer protection platform. The company is riding a wave of demand as AI-powered cyber threats make individual users increasingly vulnerable to sophisticated attacks.
Seattle Scientists Launch $95M AI Biology Effort
Seattle-based scientists have launched a $95 million AI-driven biology research initiative aimed at accelerating drug discovery and fundamental biological research. The project is one of the largest academic-industry partnerships applying foundation models to life sciences.