Confirmed
Google Launches Universal Cart Across AI Mode And Gemini
Google is rolling out Universal Cart in the U.S. inside AI Mode in Search and the Gemini app, a shared cart that holds products from multiple retailers in one place. It tracks price drops and inventory changes, surfaces promo codes and payment perks, and checks out through the Universal Commerce Protocol (UCP). The cart draws on Google's Shopping Graph, more than 60 billion continually refreshed listings, with launch retailers including Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify merchants.
In plain termsUCP is the technical standard letting an AI assistant complete a purchase across any retailer that adopts it, similar to how a single card network works at stores that never built their own payment system.
Why it mattersFor e-commerce and product-feed teams, this is a live transaction surface running on UCP outside any retailer's own checkout, worth testing for where it pulls cart share from site-direct purchases.
Our takeI wrote in 'UCP Spec Breaking Changes' that the protocol shipped four releases in eight months, three of them carrying backward-incompatible changes to fulfillment, consent, or signing fields that merchants had already integrated. Universal Cart now runs real consumer checkout on top of that same spec, so the volatility a merchant could once absorb quietly as an API maintenance chore now sits directly behind a live purchase button.
Deep dive: UCP Spec Breaking Changes: What Merchants Must Fix
Model & AI releases
AI model launches, model API price changes and major assistant features, kept separate from the search news above. Every model story Pulse has covered is collected under LLM Models & Agents, and the monthly model comparison shows how they stack up.
Confirmed
Swiss AI's Apertus 1.5 Adds Multimodal Input And A 262K-Token Context Window
Swiss AI released Apertus 1.5, in 8-billion and 70-billion-parameter sizes, succeeding the original Apertus 1.0 release. The update adds native image and audio understanding, an optional reasoning ('thinking') mode, and a context window of 262,144 tokens, four times larger than 1.0. Both sizes ship under an Apache 2.0 license with open weights, open training data, and full training details published alongside the models.
In plain termsOpen-weight means Swiss AI publishes the actual model files and the data recipe used to train them, so anyone can run, inspect, or modify the model directly rather than only calling it through a vendor's paid API.
Why it mattersFor teams weighing open-weight models for sovereign or on-prem deployment, the expanded context window and added multimodal support narrow the gap with closed frontier models while keeping the full open-data transparency few competing open releases match.
Our takeThis matters less for the specific lab and more as a marker: a context window that recently required a frontier closed model is now default in an open 70-billion-parameter release, which keeps shrinking the capability floor a smaller team can self-host.