Desk
Broad artificial intelligence coverage.
Ars Technica reports Gemini 4 Argon’s wide rollout is limited to paid tiers: AI Pro and Ultra get it in the Gemini app, but multi-threaded agent workflows in Antigravity and AI Studio require Ultra. API pricing is $2/$10 per million tokens through November 30, then $4/$20.
Google’s consumer announcement of Gemini 4 Argon positions it for long, multi-step professional tasks, citing a top Vals Index ranking, introductory API pricing of $2/$10 per million tokens and science and cyber results from partners such as Isomorphic Labs and Wiz that have yet to be independently documented.
The Verge frames Gemini 4 Argon as Google’s return to the frontier race, with Google claiming wins over GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 across coding, cyber, science, finance and law. Introductory API pricing is about a fifth of some rival list prices; the claims await third-party testing.
The Verge, citing Bloomberg, reports Epic Games told staff on an unannounced Marvel and Star Wars shooter that there would be ‘no more hand-written code’ and that art would use full AI workflows. Epic says it is embracing AI-accelerated development; Tim Sweeney says the goal is to build more, not cut people.
NVIDIA is open-sourcing the full Run:ai stack as the orchestration layer for DSX AI factories — and eight cloud and platform partners are already wiring it in.
After early DLSS 5 demos drew criticism for altering characters’ faces, NVIDIA’s new Hogwarts Legacy comparisons show subtler treatment. An October update shrinks the model about 24% to roughly 118 million parameters and cuts 4K memory use 20%. Hogwarts Legacy gets DLSS 5 on November 5.
The company now styled SpaceX Super Intelligence is combining X Premium social features, Grok chat and agents, and the Cursor coding tool into The X Subscription, with tiers at free, $8, $30, $100 and $200 a month. The newest Grok models require Premium or higher.
American Express introduced a framework that captures and authenticates a card member’s purchase intent before an AI agent completes a transaction, issuing an Intent ID and token that travel into fraud and dispute systems. Amex says it has not seen a rise in agent-related disputes and hopes the approach becomes an industry standard.
Apple’s first smart display, HomeView, pairs a 7-inch square screen with a speaker base or wall mount, Siri AI with personal context, and Matter and Thread hub functions. Personal data appears only after face recognition and a nearby unlocked iPhone confirm the user. It ships November 13.
Days before RTX Spark laptops went on sale October 16, NVIDIA announced a software slate built around local AI agents: the open Nemotron 3.5 Bolt 74B model, PAIR leaving beta, Adobe speedups, and PyTorch and Python support on Windows on Arm, along with sandboxing for agents.
In an OpenAI case study, Higgsfield AI CEO Alex Mashrabov says GPT-6 Astra-based agents extended its generated videos past ten minutes, with GPT-Image-2.5 improving readable text in images. The claims are vendor-published and best confirmed through public examples.
MIT Technology Review examines the open-model debate after Anthropic argued that Z.ai’s downloadable GLM-5.3 poses a major cyberattack risk. Both sides have commercial stakes, open safeguards can be stripped, and closed models have also been misused.