RESEARCH

A future OpenAI model is already breaking new ground

You may have heard claims that OpenAI's Astra has opened the AGI era, but OpenAI's internal model is already achieving mathematical breakthroughs. 

On Tuesday, OpenAI announced that it used an internal model "significantly more capable than GPT-6 Astra" to solve the Navier-Stokes equations, a set of problems for fluid motion and one of the Millennium Prize Problems.  

To put it simply, while the Navier-Stokes equations generally work well in determining the movement of fluids, under certain circumstances, OpenAI claims that its model proved that under certain circumstances, fluid can develop a "singularity" in finite time, meaning that a fluid's speed can grow infinitely without bound. This major implications for potential scientific breakthroughs.

In a press briefing, OpenAI said it used roughly 10,000 agents to determine that this equation breaks down, and that viscosity doesn't always keep fluid motion well behaved. While this sounds complicated, these equations explain some of the most complex phenomena in our world, including things like aerodynamics and weather forecasting, Ven Chandrasekaran, a mathematician at OpenAI, said in the press briefing. "Given that kind of enormity, understanding their fundamental nature carries very deep significance." 

However, OpenAI didn't just decide to solve this problem randomly. Sebastien Bubeck, a member of the technical staff at OpenAI, said that the company saw rumors that rival Anthropic had solved two Millennium Prize Problems, including Navier-Stokes, in which Anthropic had made more progress. Because of this, Bubeck said, the company decided to "pull together all of our compute and to put it all on that question," a decision that cost the company millions of dollars. 

Mark Chen, chief research officer at OpenAI, refuted allegations made by mathematician Tristan Buckmaster, who was working with Anthropic employee Levent Alpöge on solving the equation over the past month and claims that OpenAI took their work and finished solving it with a single prompt. "That would be a huge breach of user trust," said Chen. 

While solving this problem is a major step forward for mathematics, it has far greater implications for AI's utility in scientific and mathematical research, said Bubeck. "What happens when we are able to spend that amount of compute on problems that really matter? Developing new materials, finding cures to diseases — all of those things that we have been talking about for a long time — now they seem to be at our fingertips." 

Additionally, solving this problem may have been a stress test for OpenAI's longstanding push towards AGI, serving as an evaluation of a system that the company has been training for "general purpose intelligence," Jakub Pachocki, chief scientist at OpenAI, said in the briefing.

"This pace of progress is something to be taken very seriously," said Pachocki. "Even a week ago, we definitely were not expecting we'd be talking about a solution to Navier-Stokes today. It's a serious moment. We're developing these systems, and their capabilities are advancing faster than our understanding of them in some sense."

In an interview with The Deep View last week, Pachocki suggested a pause may be necessary to better coordinate between labs and nations.

Finding a solution to some of the most complex and foundational problems in mathematics is probably one of the most valuable use cases for powerful frontier AI. It fits directly into the utopian vision that these companies paint for an AI-powered future: systems that are finding solutions to humanity's hardest problems, such as new medicines or cures for diseases. It's also interesting to see how these labs are pushing one another by their competitive nature alone, with Anthropic being the main driver for OpenAI to want to solve this problem. While that competition could result in advances to science and medicine that benefit everyone, it's important to remember the sheer power that these companies hold in their hands, with Pachocki himself admitting that these models are advancing beyond our ability to understand them. As a society and an industry, we also have to be very careful about letting competition fuel breakneck technological advancement when we're dealing with such a potent, dangerous, and unpredictable force. 

TOGETHER WITH CRUSOE

Try Serverless Inference and Fine-Tuning, free.

Crusoe Intelligence Foundry gives you fast access to Serverless Inference and Serverless Fine-Tuning on top open models, no infrastructure to manage

Every account now includes $5 in free credits to put toward either product.

Log in and put them to work; no cluster to provision, no long setup, just a model and an API key.

Claim your $5 in credits

BIG TECH

Muse AI agent exposes Meta’s trust gap

Meta CEO Mark Zuckerberg has touted the company's commitment to building "personal superintelligence," including in a recent 6,500-word essay. Now, the company is taking its first steps in that direction. 

On Tuesday, the company announced Muse, a personal AI agent built for consumers' everyday tasks, is now available for US users. The company said that Muse can handle busywork, answer questions, browse the web, fill out firms, make purchases, create documents and more. The agent also watches user activity to proactively help with goals and make suggestions. 

In a post on X, Zuckerberg said that Muse is free to use for up to 100 million tokens per week, with subscription plans available for those who use more. Additionally, users need no technical experience to use it, Meta said, working "out of the box" for the average consumer. 

• When a user shares a goal with Muse, the agent develops a plan to coordinate their time and resources before pushing that task forward. The model checks with users before taking "sensitive actions," the company said, including purchases or sending emails. 

• However, in order for the agent to work as well as Meta purports, users must connect Muse to any platform they may use on a daily basis, including things like calendars, email accounts, payment platforms and more.  

In the company's blog post, Meta emphasized safety from the jump, claiming that Muse is built "from the ground up" to be safe, secure and private. The platform runs on Muse Secure VM, a virtual machine that holds the agent itself and a user's data. Additionally, a "sentinel" agent runs on the virtual machine, though it is kept separate from the agent at a system level. Later this year, Meta said it will launch Muse Confidential VM, a virtual machine that's encrypted with a key that only the user holds, "so not even Meta can access it." 

The model has no visibility into passwords or payment methods, and any credentials a user shares go into secure storage that allows them to use them without seeing them. Meta said that no user information is shared with its ad platform. 

The release also signals that Meta is betting big on consumer AI as rivals like OpenAI and Anthropic keep their sights set on enterprises as their cash cow. However, as one of the biggest social media providers globally, Meta has had a rocky history with consumer trust and safety, including being forced in August to pay $18 billion to settle a multi-state lawsuit alleging that the firm knowingly designed social media platforms Instagram and Facebook to addict children, and a number of lawsuits relating to its handling of private data. 

In a statement to The Deep View, Miranda Bogen, director of the governance lab at the Center for Democracy and Technology, noted that while these systems promise convenience, they require "an enormous amount of private data to function." Additionally, the consequences of agents messing up are more than just "a bad recommendation or a creepy ad." 

"With minimal input from users, these systems could take actions that damage someone’s reputation, leak sensitive medical information, or even deplete bank accounts," said Bogen. "Given the stakes, the assurances companies are offering around reliability and privacy seem concerningly inadequate."

Bogen's assertion is worth noting: The stakes of what can go wrong with a user's private data and accounts are far higher with an agent that can take autonomous action than they are with simply putting your data into a chatbot or social platform. Though Meta emphasized privacy, safety, and security, the reality of the risks and the perception of them are two different things. Meta earned a scarlet letter following the Cambridge Analytica scandal, and while people continue to use its social media platforms, its AI is already much less popular than competitors like ChatGPT and Claude. Though the company's existing consumer popularity might give it a slight advantage, given its history, the frontier labs such as Anthropic, OpenAI, and Google, or even a competitor like Apple, may have a better shot at earning the trust of consumers. 

TOGETHER WITH GENERAL ASSEMBLY

You bought the AI tools. Did you build the skills?

Most growing companies buy AI programs and expect an immediate productivity boost. What goes wrong?

They skip the training, forgetting that their people are the lynchpin of any transformation.

Leaving a busy workforce to figure out complex tech on their own quietly slows down everyday output. In fact, a third of talent leaders call this missing training their biggest bottleneck to faster execution.

General Assembly delivers the role-specific AI training teams need to actually use the tools they've paid for.

Build the skills behind your AI investment.

HARDWARE

IFA showed robots are nearing the mainstream

Robots that box, dance, climb stairs, and play soccer greeted me at IFA 2026.

While robots are typically part of the tech conference razzle-and-dazzle, this time was a bit different, as many of these robots were actually available for purchase. While the prices are steep, costing up to tens of thousands of dollars and, in most cases, not yet affordable or justifiable for most consumers, the hardware itself looks ready and has convinced me that the leap from spectacle to a product you can buy is closer than we think.

"Looking at human-like robots, it's a spectacle, it's an eye-catcher. But the question is valid: what does it mean for me? Is it just a movie, a show? Or do I benefit from it? Maybe not in the next couple of months, but in years, yes," Leif Lindner, CEO of IFA Berlin, told The Deep View. 

Lindner added that he recently visited robotics camps in Shenzhen, where one of the biggest challenges identified in making truly human-like robots was their dependence on vast amounts of data for training and operation. Also, making these household helper robots a longer-term prospect depends on consumer trust in how their data is used.

Yet AI developments will likely soon make it possible to close that gap. For instance, obtaining accurate simulated data is much easier than it was before AI exploded onto the scene, as seen in advancements by Nvidia with its Omniverse and Cosmos platforms and by physical AI startups such as Fei-Fei Li's World Labs, all of which are trying to tackle that problem. 

In terms of data, people are increasingly willing to share a lot of their personal information, data, and even real-time feeds through products like smart glasses to get smarter AI features. Ultimately, with AI helping close the gap on both of robotics' biggest obstacles, I believe a robot in the household may be a future that's closer than we think.

"Robotics is, for me, the alarm clock for everybody to showcase that innovation is here. Therefore, we are saying the future is now," added Lindner.

Of course, AI won't solve the cost problem, as humanoid robots still carry very steep price tags. Even if they reach the point where they can do everything the ideal version of them is meant to do, the purchase could be hard to justify. For instance, the Xiaomi CyberOne runs $99,999, and the Unitree G1 runs $13,500. But that's the standard curve for new hardware: computers and cars followed the same path before supply and demand evened out. And we're likely to finance robots like cars or computers. That's why I don't think humanoid robots will be commonplace in households within 5 or even 10 years. But what's striking is how fast they are advancing, with AI not just reshaping products but also how people think about data privacy. A robot-filled future, once assumed to belong to sci-fi or being decades or even centuries away, suddenly feels closer than expected.

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.

LINKS

• Google DeepMind unveils AlphaGenome Atlas, a predictive log of DNA

• Qualcomm, Amazon sign $4 billion deal for AI infrastructure

• Mistral raises $3.5 billion Series D amid data center push 

• Google Cloud, Accenture launch AI enterprise group with 1,000 FDEs

• AI coding startup Cognition raises $2 billion at $48 billion valuation

• AI healthcare startup Forus raises $150 million at $3 billion valuation

• XPENG launches automated production line for humanoid robots

• ChatGPT Images 2.5: OpenAI's new state-of-the-art image model, with sharper details, more precise editing, and faster generation.  

• Runway Team Plan: A new self-serve plan for small businesses and teams between two and nine seats.

• Krea Realtime Director: Allows users to direct high-quality video generations in real-time. 

• ComfyUI-HR-Endless-Sampler: Enables longer videos by processing them in chunks, and uses Gemma 4 to keep scenes consistent.

• Gemini Daily Brief: The proactive AI feature to help prep for your day is now available for free to all Google users who have Personal Intelligence turned on

• Nvidia: Senior Solutions Architect, Agentic AI — Safety and Security

• Deloitte: Agentic AI Engineer — Healthcare AI

• Replit: Senior AI Builder

• Docusign: AI Agentic Engineer

GAMES

POLL RESULTS

Do you think your organization is prepared for agentic cyberattacks?

Yes(6%)
Somewhat (25%)
No (63%)
Other (6%)