The arrival of the new Siri powered by artificial intelligence It's becoming one of the longest-running tech sagas of recent years. What Apple presented as a natural evolution of its voice assistant, integrated under the Apple Intelligence umbrella, has ended up being plagued by delays, roadmap changes, and internal adjustments that have sown doubt among users, developers, and investors.
Even so, the project is far from stalled. The company has made a move with a strategic agreement with Google to incorporate Gemini models In some of the smart features, it has redesigned Siri's technical architecture and continues to insist that its big leap in quality will arrive within the 2026 window, although no one in Cupertino dares to tie it to a specific version of iOS anymore.
From star prospect to recurring setback
The recent history of the new Siri begins at WWDC 2024, when Apple officially unveiled Apple Intelligence as the basis of its AI strategyAt that time, the company showcased in demonstration videos an assistant capable of better understanding context, accessing personal information under user control, and acting with much greater precision within applications.
In the following months, various leaks and analyses by specialized media placed the rollout of the revamped Siri in several phases in iOS 18 and laterBut the deadlines kept shifting. What was initially hinted at for the end of 2024 ended up being moved to 2025, and later to a generic 2026 window, without any public commitments regarding the exact version.
The latest reports pointed consistently to iOS 26.4 as the turning pointmarking March 2026 as the month in which iPhone users would begin extensively testing the new assistant. However, internal testing has revealed more obstacles than anticipated, to the point that Apple is already working with scenarios that shift key functions to iOS 26.5 and even iOS 27.
This constant shifting of dates has led some members of the public to feel they are being "dumbed down" with empty promises, although sources within the company insist that There has been no deliberate deception, but rather changes of plans on the fly, depending on how development progresses.
What does Apple want to do with the new Siri?
Beyond the dates, the underlying objective is quite clear: to turn Siri into a modern conversational assistant, with generative AI and chatbot-like capabilitieswhich leaves behind the rigidity of the current commands. The new internal architecture, known as Linwood, is based on Apple Foundation Models, now reinforced with technology from Google's Gemini team.
Among the functions under development, the following stand out: much more advanced contextual understanding, the ability to maintain the thread of a conversation during several interactions and the ability to remember relevant elements from previous conversations to give more useful and less robotic answers.
Another key pillar is deep control over the system and apps. Apple is working on a set of "application intentions" that allow for linking complex actions with a single voice command: from searching for a specific photo, editing it and sending it via messages, to locating an old email with a reservation and adding it directly to the calendar.
They are also explored multimodal and creative capabilities, such as the generation of personalized images based on Image Playground technology, or responses that combine text, on-screen context and personal data to offer summaries and recommendations tailored to each user.
Underlying all these new features is the intention for Siri to become more like... conversation models like ChatGPT or Geminibut without abandoning Apple's classic principles of privacy and ecosystem control.
Complex internal testing and phased deployment
The main reason for the delays lies in what Apple is encountering during internal testing of the new features. The development versions show a Siri that It still falls short in basic aspects such as response speed and accuracyespecially when asked to manage complex tasks or combine multiple sources of information.
Employees who are testing iOS 26.5 in a preliminary phase report queries being left unfinished, responses taking too long, and errors in interpreting what the user has said if they speak quickly. Other issues have also been detected. problems handling long requests, in which the system needs more computing time than would be acceptable for a smooth experience.
On more than a few occasions, the assistant resorts prematurely to the Integration with ChatGPT as an escape routeEven in cases where, on paper, it should be able to resolve the request using Apple's own technology. This occasional reliance on external solutions doesn't quite fit with the long-term strategy, which seeks to limit what goes outside the controlled ecosystem.
To overcome this scenario, the company is working with the idea of roll out the new Siri's features in stagesSome less critical capabilities could appear as early as iOS 26.4, while others, especially those involving intensive use of personal data, would be reserved for later versions once they are more robust.
It is possible that the first versions that reach users will include clear warnings that Certain features are in preview phase or they may not work completely reliably, a formula similar to the one Apple uses with the public betas of its operating systems.
The agreement with Google Gemini: a pragmatic shift
Alongside its internal technical advancements, Apple has decided to rely on an external partner to gain time and strength in artificial intelligence. The company has closed a multi-million dollar deal with Google to use the Gemini model family as part of the basis for the functions of Apple Intelligence and the new Siri.
The figures circulating place this pact around 1.000 million dollars per yearThis is a significant but manageable amount for a company the size of Apple if, in return, it can quickly bring its assistant's experience closer to that of its main rivals. The idea isn't to completely replace its own models, but to combine them with the maturity of Gemini where it makes the most sense.
Interestingly, public communication of the agreement has been asymmetrical. The announcement appeared clearly in the Google news channelsWhile in Apple's official spaces the reference has been much more discreet, limited to mentions by Tim Cook in conferences with investors and results presentations.
In those calls, the Apple CEO argued that Google's AI provides a very capable base for Apple modelsThis allows for accelerated development without abandoning its core design principles. At the same time, Cook has been adamant that the company has no intention of disclosing the financial details of the agreement.
For Google, the agreement is also significant: it strengthens Gemini's presence on millions of high-end devices and consolidates its role in the AI ​​market, even though it is not the company that directly controls the operating system.
Privacy and on-device processing: Apple's red line
One of the factors that most complicates the development of the new Siri is the internal requirement to maintain privacy as an essential elementWhile other players in the sector base a large part of their services on sending massive amounts of data to the cloud, Apple wants most AI operations to run directly on the iPhone, iPad, or Mac, or on its own servers designed with a private computing approach.
Executives like Craig Federighi have insisted that, when a model answers a personal question, That information should not leave the user's controlled environmentThis means preventing requests from becoming training material for third parties or from being stored in infrastructures over which the company does not have full control.
To support this strategy, Apple is developing chips specifically designed for data centersWithin internal projects like Baltra, which aim to deliver AI performance in the cloud aligned with their own silicon design, the goal is to offer advanced features without compromising on a data protection model that is stricter than the current industry standard.
In addition, the company relies heavily on synthetic data and licensed content to train its models, instead of relying primarily on data directly gleaned from customer usage. This increases costs and slows down development, but allows Apple to present a differentiated narrative against the competition.
This hybrid approach—combining on-device processing, proprietary infrastructure, and, in certain cases, support from models like Gemini—is one of the reasons why The technical integration of all the parts is proving to be so laborious.
What users can expect in iOS 26.4, 26.5 and 27
In the short term, the big question is how much of the new Siri will be visible in upcoming versions of iOS. Leaked plans paint a picture in which iOS 26.4 would serve as a gateway Some Apple Intelligence features and initial assistant improvements are coming, with a developer beta planned for late February and a public release around late March or early April, subject to change.
From there, iOS 26.5 would act as an intensive testing platform For employees and internal testers, this includes new Siri features explicitly marked as experimental or in preview. It's possible that certain capabilities will be rolled out to a small group of users first to gather feedback before a full-scale deployment.
Looking ahead iOS 27, scheduled for fall 2026 in parallel with the launch of the new iPhonesExpectations point to a Siri closer to a classic chatbot. This assistant could offer a revamped interface, greater system control, deeper management of files and personal data, and clearer integration with Apple services on macOS and iPadOS.
However, everything points to the fact that Not all iPhone models will have the same featuresThe most demanding AI functions could be limited to recent generations —along the lines of what is already being seen with Apple Intelligence—, which in practice would limit the full experience of the new Siri to a portion of the device fleet.
Although the timelines keep shifting, the feeling in the industry is that Apple doesn't want to prolong much longer a project that is already almost two years behind schedule compared to what was initially suggested.
Whatever the final schedule, the combination of a revamped architecture for Siri, support for models like Gemini, an obsession with privacy, and a phased rollout in iOS 26.x and 27 This paints a picture where Apple's assistant could shed its image as a stagnant service. It remains to be seen whether the company can square the circle between technical ambition, market expectations, and its own internal quality standards, but all indications are that the next big move in AI assistants will largely revolve around this new generation of Siri.