Stay ahead on AI.
All in ONE place.

The Claude Code desktop app now supports custom themes via mods, allowing users to change colors and add animated backgrounds. Users can describe their desired look directly to Claude or get started quickly using the theme-examples plugin. Themes currently work in local sessions only, with cloud session support coming soon. Users are encouraged to update to the latest version of the desktop app to try the feature, and documentation is available for those who want to write themes manually.

Anthropic has expanded access to Claude Duet, inviting all US-based Pro, Max, and Ultra subscribers who were on the waitlist as of the previous day. The research preview includes usage limit resets: three for Pro subscribers and one each for Max and Ultra subscribers. Users should check their inboxes for invitations. Anthropic says it aims to expand Claude Duet to countries outside the US very soon. Those not yet on the waitlist can still join via the provided link.

Research by former OpenAI scientist Will DePue of Unemployment Labs found that top Chinese AI models from DeepSeek, Moonshot AI, MiniMax, and Ant Group were trained on Nvidia Blackwell GPUs banned from export to China. DePue analyzed model weights, which retain fingerprints of training hardware, to identify Blackwell chip signatures. Epoch AI estimates roughly two million advanced chips entered China in 2024-2025 via smuggling schemes, with 275,000 Nvidia accelerators imported to Malaysia alone, a common workaround for Chinese firms.

Anthropic's William Hu published a guide on building always-on AI agents using five core primitives: an event inbox, timers, memory, permissions, and an outbox. Unlike typical agents triggered only by user messages, always-on agents are stateful and proactive, waking periodically to handle webhooks, scheduled tasks, and background jobs. The guide notes recent model advances have made these agents more practical, with Opus 5.5 costing 78% less per solved task than Opus 5 on long-horizon coding benchmarks.

OpenAI has launched ChatGPT Wallet, a secure payment system for AI agents built in partnership with American Express, Mastercard, and Visa. The wallet lets agents complete purchases without exposing users' actual card details, requiring biometric security checks like Face ID or fingerprint scans for every transaction. Initially available to ChatGPT Pro subscribers in the US through the dots feature, it preserves existing card rewards and dispute protections. OpenAI also released a Commerce Partner Playbook to help businesses support agentic purchases through existing systems.

Google has added Google Wallet to its Gemini app's Personal Intelligence feature, rolling out now for all U.S. users on web, Android, and iOS. Users can connect Wallet and optionally link bank and credit card accounts via Plaid for deeper insights. By tagging @Google Wallet in chat, users can ask Gemini to visualize spending, find forgotten subscriptions, and build holiday budgets. Gemini can only read account data and cannot move or invest money.

Reuters reports that ousted Nvidia CEO Jensen Huang had a relationship with a third employee, adding to two previously reported extramarital affairs that led to his resignation. The other two women are Alice Huang, suspected by US intelligence of being a Chinese spy, and Kimberly Powell, Nvidia's VP of healthcare. CFO Colette Kress was named interim CEO. Nvidia's board is searching for a permanent replacement as the company's stock has fallen over 30% since its early October peak.

Google has launched Gemini 4 Hydrogen, a fast model built for everyday coding and agentic workflows, priced at $1.50 per million input tokens and $7.50 per million output tokens across a 1 million-token context window. Available in Google AI Studio, the Gemini API, and the Gemini app, Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash and triples Terminal-Bench 4.1 scores. Google also introduced Orchestrator in Google Antigravity, which dynamically switches between Hydrogen and frontier models, achieving 19% faster completion with 24% less quota usage.

Google has launched Gemini 4 Hydrogen, a cost-efficient model designed for everyday coding and agentic workflows. Built for rapid iteration, it handles repository coding, debugging, interactive web development, and multimodal workflows. It runs standalone across four configurable thinking levels or alongside frontier models like Gemini 4 Carbon and Gemini 4 Argon. Hydrogen features tighter instruction following and fewer fabricated claims. It's available via the Gemini API, Google AI Studio, Android Studio, and for paid AI Pro or Ultra subscribers through Google Antigravity.