Behind closed doors at Apple, Siri is being reconsidered that could change the way we talk to our devices. Apple has been quietly working on a ChatGPT-like app to help its engineers test the revised version of Siri. The app won’t be released to the public and is strictly for internal testing, but it offers a rare glimpse of where Siri is headed.
The test bench points to something bigger. Apple knows that Siri’s current architecture is out of reach in a world of conversational AI. The app is being used to test new Siri features, such as greater contextual awareness, the ability to do more within and across apps, and deeper integration with personal data. Think about taking a shopping list in Notes, booking a delivery window in a shopping app, and then placing reminders on your calendar without losing track. Not just better small talk, smarter workflows.
What’s really going on behind the scenes?
Here’s the timeline: Apple has been working on a smarter version of Siri since last year. The company initially planned to introduce an Apple Intelligence Siri as part of the update, but delayed the functionality until 2026. The initial architecture gave way to the context and demands of the conversation, so Apple paused and rethought the foundation.
Apple couldn’t get the first-generation Siri architecture it was using to work properly, so Siri needed to be rebuilt from the ground up using a second-generation architecture that relies on large language models. That meant redoing the conversation management, the long-term context, and the framework that ties into the apps. The Apple Intelligence Siri plan was reworked and Apple decided to renew Siri with a second-generation architecture, accelerating the transition to an LLM.
The internal testing application is the laboratory bench. Basically, the app takes the still-in-progress technology of the new Siri and puts it in a form that employees can test more efficiently. It also helps Apple evaluate whether a chatbot-style interface has value, allowing the company to gather feedback on whether the chatbot format has value. Faster iteration, clearer signals, less guesswork.
The LLM transformation comes to your pocket
Simply put, the next Siri won’t feel like the Siri you know. Apple is testing and evaluating models from vendors such as OpenAI, Anthropic, and Google (Gemini), along with internal models. As a result, LLM Siri will be able to maintain ongoing conversations, provide human answers to questions, and complete more complex tasks.
The test application points out how that plays out. It allows Apple users to manage multiple conversations on different topics and can save and refer to previous chats, keep track of previous queries, and support long back-and-forth exchanges. Plan a vacation and then return days later to swap flights and modify the hotel. Start a research thread on iPhone, pick it up on Mac, and then check it all out when you add a reminder. No more Groundhog Day with every consultation.
That memory becomes sharper when combined with Apple’s “World Knowledge Answers” system. Apple doesn’t just clone a chatbot. They’re building World Knowledge Answers internally, and some Apple executives refer to it as an “answers engine.” It will include an interface that supports text, photos, videos and local points of interest, with Siri able to summarize search results to provide a clear summary of the content.
That positions Siri against Google’s AI overviews and tools like Perplexity. Not just timers and lights anymore. A center that can understand stratified questions in all formats and provide you with a useful answer.
The conundrum of partnerships: who drives the future?
Strategically, Apple is playing big. Apple has held talks with Anthropic, OpenAI, and Google, and may use an outside company to power the new version of Siri rather than relying solely on its internal AI models.
Tellingly, Apple is reportedly testing a custom model of the Google Gemini under a deal, which could then power some of Siri’s features. At the same time, Apple is still evaluating models designed by Anthropic and its own internal models for planner functionality, and they could use Google’s models there as well.
This combination of multiple vendors allows Apple to align the strengths—Google’s search knowledge, Anthropic’s reasoning, and Apple’s privacy-focused data work—without giving up control of the experience. Within that setup, Apple’s Foundation Models will be used to search for user data, so sensitive information stays in-house.
Pragmatic, not precious. Build what matters internally, borrow what accelerates the roadmap, keep user experience and privacy intact.
When can you really use this?
Here’s the roadmap: Apple is reportedly aiming to release the LLM version of Siri in early 2026, likely as part of an iOS 26.4 update that could arrive in March. That comes about a year later than the original target, a full year after Apple originally wanted to release the updated version of the personal assistant.
More waves are queuing up. Apple plans to introduce a new look for Siri late next year, giving it a more humanoid design similar to the Mac Finder logo. Then, later in 2026, Siri will receive a visual redesign and a built-in health feature that will be the backbone of a paid wellness subscription service.
The search will appear first in familiar places. You’ll see it in Siri, Safari, and Spotlight, making them touchpoints for Apple’s answer engine.
Yes, that cadence puts Apple behind faster-moving rivals. It also gives them time to make the experience feel native, not forced.
What this means for the Apple ecosystem
This is not just “Siri, but smarter.” It’s Apple trying to make conversation the glue between tasks. The current ChatGPT hook within Apple Intelligence seems like a guest appearance. LLM Siri is Apple taking the microphone.
There is a twist. Apple executives have expressed doubts about the usefulness of chatbot interfaces for artificial intelligence functions. Craig Federighi says Apple wants to create an experience that is “integrated into everything you do, not an extra chatbot” at WWDC. This makes internal testing an eye-opening move. Apple is checking whether a classic chat window holds up, even when they first design for the integration.
If it works, productivity changes shape. You have a single conversation that spans research, planning, execution and monitoring. Less app juggling, more momentum. That’s the kind of polish people buy the ecosystem for.
Where do we go from here?
The in-house ChatGPT-like app shows that Apple is getting serious about the assistant race. Apple has reportedly scrapped its plans to launch a separate chatbot app and opted to incorporate AI into its native search layers, but they’re not closing the door on conversational interfaces.
The 2026 window seems far away when others ship monthly. Still, the method is familiar: test quietly, shape the edges, and run it when you feel like Apple. Can they maintain privacy and tight integration while outperforming their rivals in terms of capability? That’s the ball game.
If you live with Apple devices, this could be the biggest change since the iPhone era. A Siri that remembers, understands context, and moves between apps with you. The assistant that Apple hinted at more than a decade ago is finally within our reach.
Image via Apple