🏆 Google's Gemini 4 Beats Opus 5.5 and GPT-6 Astra

Google has given us a first look at Gemini 4 Argon, and its early results suggest it can challenge OpenAI and Anthropic on demanding work. Most of us can’t try it yet, but there’s a reason to take another look at what Google is building. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Google has unveiled Gemini 4 Argon, a new AI model designed to work through longer, more demanding tasks in coding, finance and cybersecurity.

Google is using Gemini 4 Argon to tackle large engineering projects across its products and infrastructure. The company reports that its agents helped free over 300 TiB of data-center memory through software improvements. Teams are also using it to rewrite large codebases, with automated testing and human review before deployment. Broader access will start with paid API customers and Google AI Ultra subscribers after further testing and safety work, though Google has not set a release date.

Google’s reported results give it a stronger case for winning work that people currently give to ChatGPT and Claude. Its reach through Search, Workspace and Cloud makes that competitive push significant. If Google brings these capabilities into those products, users could have more reason to handle complex projects within tools they already use. Wider access will show how well Gemini 4 Argon performs outside Google’s own examples. To turn this into lasting adoption, Google will need to help people hand over larger parts of a project while spending less time guiding each step and correcting the results.

Evolving AI: Connecting cost, performance and infrastructure so you can scale responsibly.

AI spend is growing fast and for most organizations, it’s growing in the dark. This practical guide is for engineering, finance and FinOps leaders who need more than dashboards in the age of AI; they need answers.

Inside, you’ll learn how to:

See how Kevel cut AWS costs by up to $100,000 per month after replacing reactive cost reviews with real-time visibility enabled by Datadog.

Free Download →

Evolving AI: OpenAI, Anthropic and other AI labs are under investigation as the FTC examines whether their technology puts people at risk.

The FTC is looking into whether AI companies have treated consumers unfairly or misled them about their products. The probe comes amid reports of agents accessing outside systems without permission, including OpenAI agents breaching Hugging Face, where developers share AI models and code. Investigators are preparing “civil investigative demands,” similar to subpoenas, which legally require someone to provide documents or testify. The investigation was already underway when President Trump and AI leaders signed their voluntary safety pact. No finding of wrongdoing or penalty has been announced yet.

OpenAI and Anthropic may have to show regulators how they kept their agents under control and what they did when those controls failed. That could help establish whether their actions matched their safety promises. And you don’t have to use ChatGPT or Claude to have a stake in this, because an agent could reach a service that holds your information. If your data is accessed without permission, you should hear about it quickly and know who is responsible for fixing the problem.

Claude is helping the IRC turn messy frontline health data into decisions much faster, with one pilot cutting analysis that took a full working day to under 30 minutes. This short film shows why that matters in Nigeria, where nutrition workers still record child health data by hand and slow processing can leave valuable information unused. It is a powerful look at AI helping aid teams spot needs earlier and stretch limited resources further.

Evolving AI: The Pentagon has launched Project Meridian to study how AI could change warfare and help decide which technologies the military should develop.

The Pentagon wants the group to look beyond today’s conflicts and consider what troops might need decades from now. Its technology chief, Emil Michael, will oversee the study, drawing on military research and outside expertise to recommend which systems to develop and test. It also plans to establish a separate Autonomous Warfare Command to help military branches test and deploy drones and robotic systems faster. The target is October 2027, though congressional approval is still needed.

The Pentagon is giving people whose companies supply the military a say in what it should build next. Their experience could help it make better decisions, but their companies could also benefit from the recommendations, which makes independent review important. Earlier programs such as Replicator and Drone Dominance have already helped put uncrewed systems into service. A dedicated command could give that work clearer leadership across military branches. Closer ties between troops and developers could also help soldiers get useful equipment sooner and have problems fixed more quickly when their needs change.

QUICK HITS

💻 OpenAI and Synopsys are building a specialized chip-design model that will learn to use Synopsys tools across semiconductor design workflows.

🧬 Anthropic built a robot exposure index estimating robots can technically perform about three-quarters of physical tasks in U.S. jobs.

🧪 Google DeepMind introduced SynthID Bio, extending its watermarking work to help identify AI-generated synthetic biology designs.

👓 Meta began rolling out its Wearables Toolkit 1.0, letting developers connect mobile apps to cameras, microphones and displays on its AI glasses.

 *partner link