GOVERNANCE

OpenAI's GPT-6 Astra is here. Is the world ready?

In the shadow of the Hugging Face security breach, OpenAI's most powerful model is about to be loose in the world, but with new safeguards.

On Thursday, OpenAI unveiled GPT-6 Astra, its next-generation model that it characterizes as a "generational leap in capability." Starting today, it will be coming to a limited set of organizations, including those in the Daybreak Access program and then rolling out to ChatGPT customers on Plus, Pro, Business, Enterprise, API, and AWS "over the coming days," according to the company.

In a briefing with the press, OpenAI President Greg Brockman said that with Astra, "It's not unreasonable to feel that we are now in the AGI era," and that the release is just the beginning. "I feel like there is a qualitative shift we've gone through," said Brockman. "I think that it is a significant moment, but it is more about the continuum than it is about any individual point along the way." 

Notably, the API for the new model will cost $10 per million input tokens and $50 per million output tokens, or the exact same price as Anthropic's Mythos/Fable 5 and 5.1, making Astra one of the most expensive models on the market. Still, OpenAI emphasizes that Astra is able to use "substantially fewer total tokens per task" in multiple scenarios, directly targeting the efficiency that enterprises are after. 

As for its capabilities, OpenAI noted a number of improvements that Astra offers over its predecessors and competitors:

• The company has called Astra "the world's best computer use model," with the best speed, accuracy and safety on the market, and capabilities in a number of domains.

• The model is also its best yet for professional work and makes leaps in scientific discovery, mathematics, and health research. 

• OpenAI also says Astra is "the best model for software engineering to date," outranking GPT-5.6 Sol and Claude Fable 5.1 on DeepSWE v1.1, a benchmark for complex software-engineering tasks.

Given the model meeting the "critical" threshold for cyber capabilities under its preparedness framework, the company said that Astra will refuse to comply with advanced cybersecurity tasks, such as exploit discovery, and features stronger protections against cyber misuse. The company also said Astra is its "most aligned model," with improvements in respecting task boundaries and transparent communication, and is three times less likely than GPT-5.6 Sol to inaccurately represent its capabilities. 

However, OpenAI found that Astra’s written reasoning is harder to monitor than GPT-5.6 Sol’s. Jakub Pachocki, chief scientist at OpenAI, said in the briefing that as these models become more intelligent, "monitorability is getting more challenging." This is because the smarter a model becomes, the less language reasoning it needs to be able to complete tasks.

"We see monitoring is critical, and we take this trend seriously, and we believe monitoring is still a very core technique for Astra, but for future models improving it … is a research priority," said Pachocki.

Despite the industry's push for efficiency and affordability, OpenAI has now put itself into a position where the market expects it to constantly release something bigger and better than before to be worth that trillion-dollar price tag. And while OpenAI's mission is to democratize intelligence, the reality is that Astra is not meant for everyone. It is best suited to handle the most powerful and critical tasks that organizations have to offer. Embedding itself within those workloads with increasingly powerful models like Astra could make its AI a foundational part of some of the most important work that's being done within enterprises, making safety vital. And after the Hugging Face incident, there will be an even greater microscope on the model. 

IN PARTNERSHIP WITH LAMBDA

Struggling with MFU? Steal this framework

Whether you're an ML engineer deep in the trenches or a manager focused on getting the most out of your time and money, one thing rings true: efficiency matters. As it turns out, most teams out there are using less than half the computing power they're paying for, with no idea how to improve, which is where Lambda comes in.

The reproducible framework Lambda built boosts MFU by over 25% without changing the model itself. The entire framework is in this whitepaper, which also shows you how to address the memory inefficiencies inflating your costs, the training configurations that aren't making full use of your hardware, and the bottlenecks slowing down your GPUs

Download Lambda's whitepaper and fix your MFU today.

PRODUCTS

Nvidia closes the Windows-Mac local AI gap

Mac has been the undisputed home of local AI, but Nvidia is giving Microsoft an assist to get Windows in the game.

On Thursday at IFA 2026, Nvidia unveiled new tools and hardware aimed at making it easier for more PC users to run AI locally on Nvidia GPUs for faster local inference. For starters, Nvidia RTX Spark arrives in October in new Windows PCs from Lenovo and Acer, on display this week at IFA. RTX Spark has a 1 Petaflop RTX Blackwell GPU, up to 128GB of unified memory and a highly efficient 20-core Grace CPU, which, when combined with new agent frameworks, gives Windows PCs the power to run always-on AI agents locally. 

Nvidia is also offering a simpler model setup on Windows for three of the most popular agent apps: Perplexity Portable Computer, Hermes Agent, and OpenClaw. Here is a quick rundown of the features, according to the release: 

• Perplexity Portable Computer: Will be available on Nvidia RTX GPUs with at least 24GB VRAM running Linux or Windows

• Hermes Agent: Coming soon, configuring a local model in Hermes will be streamlined with one-click setup across RTX and DGX systems on both Windows and Linux 

• OpenClaw: Nvidia worked with Microsoft to reduce set-up friction and the result is that the OpenClaw Windows App simplifies setting up an optimized local model on any RTX GPU with at least 24GB of VRAM

Building on its efforts to improve AI use efficiency, the tech giant also unveiled NVIDIA Personal AI Router (PAIR), a free, open-source software tool that can coordinate a household's PCs to run local AI together. NVIDIA says that PAIR can automatically discover compatible PCs on a local network and route independent inference requests to the system with available capacity. Ultimately, this is meant to bypass the bottleneck that is caused when multiple agents or tasks are waiting on a single GPU.

The rise of AI agents has led working professionals to discover brand-new ways AI can assist with their work. However, the caveat is that agents can quickly rack up token costs and so power users typically want to move to running AI locally. Until now, it has largely meant a reliance on Mac products, as seen by the shortage of the Mac mini  and Mac Studio, and the rollout of cutting-edge AI features arriving on Mac first. The release of Nvidia RTX Spark on Windows by Nvidia is significant, as it gives users who don't want to be locked into Apple's walled garden more choice. It may also sway developers from overlooking Windows when unveiling the latest features on desktop apps, which would be a win for Microsoft and for Windows users. 

Disclaimer: Sabrina Ortiz's travel to IFA 2026 was paid for by IFA. The Deep View's coverage is editorially independent from the companies we cover.

TOGETHER WITH OUTSKILL

Master the 15+ AI Tools That Actually Matter (Live This Saturday)

205,000 jobs were cut this year with AI cited as a reason. The people who stayed knew which tool to use for what. AI layoffs have matched all of 2025 in under eight months, and AI is now the top-cited reason for corporate job cuts.

Workers with AI skills are earning a 62% wage premium, with demand growing 8x faster than the market. The split isn't people versus AI, it's people who use it well versus people who don't.

This 3-Hour AI Tools Workshop fixes that: the 15 most powerful AI tools of 2026, and which one to use for the job.

• Pick the right tool instead of one app for everything

• Build apps and dashboards with zero coding

• Wire tools together so work runs itself

This Saturday, 10 AM EST | 3 hours live + bonus session | Worth $395. Free for readers.

Save my free seat

MARKETS

Hugging Face chooses Nvidia to scale open models

Nvidia has officially acquired Hugging Face, and the two companies say there are two reasons the deal makes sense.

After weeks of reports about an impending tie-up, Nvidia and Hugging Face made things final on Thursday, announcing that the world's leading AI chipmaker had purchased the open-source AI platform in a $12.9 billion acquisition. It's Nvidia's second-largest acquisition ever, after its $20 billion purchase of Groq in December. 

Naturally, the deal has spurred fears about centralization of power and resources, since Nvidia is one of the handful of big winners benefiting from the current AI boom. Open models are viewed as one of the ways to counterbalance AI power being concentrated in a few US corporations. In a briefing with the press, both Nvidia and Hugging Face went to great lengths to assuage those fears, since the future trajectory of Hugging Face depends on earning and retaining the trust of developers and AI builders at the grass roots across the global tech ecosystem. 

But both companies emphasized that Hugging Face being part of Nvidia has two benefits:

1. It gives Hugging Face access to compute that it needs so users can keep experimenting with training, fine-tuning, and customizing open models.

2. It allows Hugging Face to scale to a lot more users in the years ahead when open models are poised to play a much larger role in the AI ecosystem. In order to get there, CEO Clem Delangue said on X that Hugging Face "needs more compute, more support, more collaboration and more visibility." Today, Hugging Face has 200,000 companies using the platform and its goal is to 5x in the next few years.

In his blog post about the deal, Nvidia CEO Jensen Huang addressed the elephant in the room about whether Nvidia would try to manipulate the open model ecosystem to serve its commercial purposes. "Hugging Face will remain an open platform for the entire AI ecosystem," wrote Huang. "Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want. NVIDIA compute will not be required to build on or deploy through Hugging Face."

Delangue also doubled down on Hugging Face's grassroots mission, saying it's not good for the AI ecosystem to consolidate around a few proprietary models. In the press briefing, he said, "We allow AI to be more distributed throughout the world and avoid too much concentration of power."

It's noteworthy that Hugging Face reportedly turned down a $500 million investment by Nvidia that would have valued the company at $7 billion in late 2025 because it didn't want a single investor to have too much sway over its decision-making. What's changed since then? The AI ecosystem has accelerated dramatically, with the pace of AI models quickening, venture funding increasing, and a new wave of open models launching. Hugging needed a reliable scaling partner, and Huang's leading role in July in calling for the AI industry and the US government to support open models likely helped give Hugging Face confidence in Nvidia's open-source stance. While I believe both companies are sincere in their rhetoric about open models, there's still long-term concern that there will be gravity pulling the platform into being a revenue driver for one of the world's most valuable public companies. And that could naturally lead Hugging Face toward choices it wouldn't make if its top priority was to simply provide an open ecosystem for AI builders.

LINKS

• How HTC Vive Eagle fixes two flaws in AI smart glasses 

• G20 nations back US framework for lighter-touch AI regulations

• Parents turn to AI apps to manage household schedules and tasks

• PlusAI to go public via SPAC merger at $800 million valuation

• Claude, ChatGPT and Grok face widespread outage on Thursday

• OpenAI commits $1 billion to model access, training for essential services

• MAI-Transcribe-2: Microsoft launches top-speed transcription model

• WeatherNext 3: New Google DeepMind, Google Research weather forecasting model 

• Claude: Gains background computer control for multitasking in Claude Cowork and Claude Code

• Muse Spark 1.3: Rolling out in Muse Code and Meta Model API in what Zuckerberg calls "frontier performance almost too cheap to meter"

• Nvidia: Solutions Architect, AI Models

• Google: Research Scientist, Cloud AI Research

• The Center for AI Safety: Research Manager

• TikTok: AI & Data Architect - Trust and Safety

GAMES

POLL RESULTS

Do you think Google should continue to lean into affordability and efficiency?

Yes (84%)
Somewhat (12%)
No (3%)
Other (1%)