AI Agents & Supercomputing
Google Expands Gemini Tasks as NVIDIA Unveils Rubin HPC & Proton Lumo
Google rolls out free agentic task execution on Gemini 4 Hydrogen across more than 60 connected applications while NVIDIA introduces Nemotron for Science at SC26 and Proton launches encrypted Lumo Desktop.

Your listening Queue & catch up
Add a story or build a catch-up queue to start listening.
Artificial intelligence platforms expanded their autonomous task execution, scientific simulation hardware, and local desktop privacy controls on Tuesday as Google, NVIDIA, and Proton each introduced major architectural updates.
Google Rolls Out Gemini 4 Hydrogen Task Execution and AP2 Payments
According to the Google AI blog, Google has begun rolling out upgraded task-completion capabilities in the Gemini application free of charge for hundreds of millions of users across the United States, Japan, Korea, Canada, Australia, and India, powered by the company's new Gemini 4 Hydrogen model. Advanced task features are also launching first in the United States, enabling users to switch the home-screen toggle from Chat to Task, forward emails to gemini@google.com, or text Gemini over RCS.
The update connects Gemini with more than 60 popular apps through Google's Connected Apps Platform, integrating with Google Calendar, Google Maps, Gmail, Google Password Manager, Google Shopping Graph, Google Pay, Google Wallet (which is expanding to iOS starting Tuesday), and third-party services such as Instacart. In documented examples, Gemini placed simultaneous phone calls to two dealerships for a 2019 Honda Odyssey service appointment, disclosed immediately that it was an AI, scheduled a Friday morning slot in under five minutes, analyzed utility statements to diagnose a $389 electric bill caused by a heat pump activating at 4 p.m. during peak pricing, and applied a $10 discount code on a USB-C charger purchase.
For commercial transactions, Gemini utilizes the Agent Payments Protocol (AP2), working with Mastercard, Visa, and American Express (coming soon) to secure checkouts with digital tokens while preserving cardholder points, airline miles, cashback, and upfront spending caps, ahead of a planned integration of the Universal Commerce Protocol (UCP). Each assigned task runs inside Google Cloud's Agent Compute Engine within a dedicated, hardware-isolated virtual machine tested by security red teams, paired with prompt-injection filters, Google Safe Browsing restrictions, and Google's AI Vulnerability Reward Program (AI VRP).
NVIDIA Unveils Nemotron for Science and Rubin HPC GPU at SC26
At the SC26 conference in Chicago on Monday, Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA, introduced new agentic scientific computing software and hardware, as reported by the NVIDIA Blog. Looking back 20 years to his first public talk on CUDA at SC06 in Tampa, Florida, Buck noted that NVIDIA now provides more than 1,000 CUDA-X libraries and emphasized that simulation remains the foundation of scientific computing alongside AI, quantum systems, and software agents.
Buck announced Nemotron for Science, which pairs the open Nemotron 3.5 Super VL model with an open chemistry dataset and post-training recipes for life sciences and chemistry that researchers can run locally on a DGX Spark, workstation, or supercomputer. At Argonne National Laboratory, an agent built with the toolkit planned and tested hypotheses from a single prompt and identified two new binding sites. Meanwhile, a KTH Royal Institute of Technology agent repaired broken GROMACS setups and fixed 81% of test cases in one overnight run on a single DGX Spark; a SETI Institute agent at the Allen Telescope Array plotted a methanol line and retuned for Doppler shift between NGC 869 and 896; and NVIDIA's Nsight AI Porting Agent accelerated a Weather Research and Forecasting (WRF) model routine by more than 5x using roughly 150 million tokens.
On the hardware front, Buck unveiled the Rubin HPC GPU within the Vera Rubin platform—expected in 2028—featuring FP64 through FP4 precision support, sparse-data acceleration, seventh-generation NVIDIA NVLink, and more than 50,000 CUDA cores, or roughly three times as many as Hopper. He also noted that the German Climate Computing Center (DKRZ) selected Vera CPUs for its HLRE-5 supercomputer expected in early 2028, joining Texas Advanced Computing Center deployments, while Lambda deployed NVIDIA Photonics co-packaged optics switches that are 15x more reliable on bit errors and link flaps. In quantum computing, NVIDIA, IBM, and Fermi National Accelerator Laboratory announced a co-design platform pairing NVIDIA DGX B200 systems with IBM Quantum Nighthawk processors, alongside magic state cultivation on Jülich's JUPITER supercomputer that cuts logical-step operations tenfold and a CUDA-Q decoding algorithm, TCBP-JOBS, aimed at cutting decoding latency from 300 microseconds to 10 microseconds and reducing error-correction computation from 12 weeks to 45 minutes.
Proton Launches Encrypted Lumo Desktop App for Windows, macOS, and Linux
In consumer and enterprise privacy software, TechCrunch — Artificial Intelligence reported that Swiss privacy company Proton launched Lumo Desktop on Tuesday for Windows, macOS, and Linux. The desktop client enables local task automation across file-system access, shell execution, browser content fetching, Proton Mail, Confluence, Dropbox, Jira, and local or remote Model Context Protocol (MCP) servers.
Proton allows users to configure granular per-connector permissions—specifying which actions execute automatically, which require explicit approval, and which are blocked—including per-task permissions for scheduled automations and reusable Skills templates migrated from Custom Lumos. Proton encrypts chat histories so only the user can read them, keeps no server-side logs, and hosts data under Swiss privacy law on European infrastructure outside the scope of the U.S. CLOUD Act, backed by its first NVIDIA B300 cluster built on Open Compute Project (OCP) standards now going live in Switzerland.
Sources
Follow the coverage
Published reporting connected by NVIDIA.
- NVIDIA Unveils Rubin HPC GPU and Sweeps MLPerf as Lambda Deploys CPO
- NVIDIA Details Supply Chain as Apple Tests Siri Caps & OpenAI Grows
- UC Berkeley Joins Schmidt AI Network as NVIDIA Launches DGX Spark Kit
- Google Expands Gemini Tasks as NVIDIA Unveils Rubin HPC & Proton LumoYou are reading this story
Comments (0)
No comments yet. Be the first to share a thought on this story.
Request a correctionContact the newsroom