Apple picks Gemini for Siri and opens the iPhone to three models

With iOS 27 any user will be able to choose Claude, ChatGPT or Gemini as the engine of Apple Intelligence; for Latin America, Gemini will arrive by default on millions of iPhones in the fall.

Generated automatically · sources linked · no prior human review

Apple closed the largest AI deal in mass-market consumer hardware: Gemini, Google’s model, will rebuild Siri from scratch in iOS 27. The announcement came at WWDC 2026, Tim Cook’s last keynote as Apple’s CEO, with striking numbers: a custom model with 1.2 trillion parameters at a cost of approximately $1 billion a year.

The decision does not close off the ecosystem. At the same time, Apple presented an Extensions system that, for the first time in iOS history, lets any user choose between Claude, ChatGPT or Gemini for Apple Intelligence features. The three big frontier AI labs on the same device: a shift that moves the field of competition. It is no longer just about being “the model,” but about being the extension the user turns on by default. For Latin America, where the iPhone holds between 30% and 40% of the premium segment in the five largest markets, the concrete effect is that Gemini will arrive as a preinstalled option on millions of devices in fall 2026, while Claude and ChatGPT will be available but will require explicit activation. None of the three models incorporated a regional perspective into its design: the region was not consulted.

Also today

In the region

Colombia is 13 days away from its presidential runoff (June 21) with the third documented AI deepfake of a candidate in the same campaign. The fake image of Espriella alongside Arias (designed to erode his image among center-right voters) adds to an earlier poll bearing the Gemini logo and to emotional AI-generated campaign videos. The Foreign Ministry rejected foreign interference, but without technical and legal instruments to act quickly. The contrast with Brazil is the sharpest in the region: that country approved PL 2338 on May 27 and its Superior Electoral Court has five active measures for the October elections. Colombia votes without that scaffolding. In Europe, meanwhile, the AI Act clock shows 55 days until the obligations for high-risk systems take effect with real power to fine.

Launches

  • iOS 27 Developer Beta 1 — Siri with Gemini and multi-model Extensions — The first mainstream mobile operating system with native AI model selection. Available starting today for registered developers; the public version is expected in the fall.

  • grok-imagine-video-1.5-preview — A new image-to-video model from xAI, up to 720p, with motion control described in natural language. Accessible directly via API and without a premium subscription, which makes it a low-cost option for experimenting with video generation.

Threads we’re following

The story of Colombia and electoral deepfakes has been on this blog for weeks. Since late May we have documented that Law 2502 was approved without operational enforcement mechanisms for the electoral period: a clock that keeps ticking. Today that scenario is confirmed with a third documented deepfake, 13 days before the vote. The next chapter is the June 21 runoff, which will be one of the most concrete cases the region has on whether it is possible to defend a presidential election from generative AI without effective regulation in place.


If Apple decided to pay $1 billion a year to put Gemini at the core of Siri and still opened the door to Claude and ChatGPT, are we seeing the end of the race to be “the model” and the start of the race to be “the preferred extension”? And what does that mean for the 400 million users in Latin America who are unlikely to pay for a premium subscription to any of the three?

About this entry. It is generated automatically from public sources, without human review before publication. It may contain errors of interpretation or summary; please check each story against its original source (the links lead there) before citing it or making decisions based on it.

Doble Click is written with Anthropic models.

Spotted an error? Report it

Tell us what's wrong, quoting the sentence if you can and, if you have it, the source that corrects it. An automated process reviews reports every night: if the error is verified, the page is corrected and a correction note is added at the bottom.

Your email is optional: we only use it if we need more context about the report. It doesn't subscribe you to the newsletter.