Apple tests one-shot siri prompts to match gemini and chatgpt
Siri is about to swallow a 1.2-trillion-parameter pill. Apple has quietly begun stress-testing a version of the assistant that can juggle multiple commands in a single breath—something Google Gemini and OpenAI’s ChatGPT already do while Siri is still asking “Which app?” for every micro-task.
The internal builds, confirmed to Bloomberg’s Mark Gurman by engineers who demanded anonymity, let users string weather checks, calendar inserts and outgoing texts into one sentence. The pipeline is slated for iOS 27, iPadOS 27 and macOS 27, according to the same people, with a first public glimpse expected nine weeks from now at WWDC on 8 June.
Why the hurry now
Apple’s cloud-side Siri currently limps along on 150 billion parameters. The coming overhaul—fueled by Apple’s $1 billion-a-year licensing of a custom Google Gemini model—lifts that count to 1.2 trillion, an eightfold jump that finally gives Cupertino the same transformer muscle its rivals have flexed since 2022. Translation: cross-app workflows, on-screen awareness and a chat interface that no longer feels like a 2014 throwback.
The catch? Nothing ships with iOS 26.5 Beta 1. Apple prefers to keep marquee AI for major-number releases, so unless the company breaks form, the new Siri brain stays locked until the 27 cycle. Early adopters who flash the first 27 beta this summer should expect the usual minefield: dropped calls, battery vampires and the occasional refusal to acknowledge that you’re holding an iPhone at all.
What changes for the user
Imagine landing at JFK and barking, “Tell Mom I’ve landed, queue Work playlist, calendar dinner tomorrow 8 pm Keens, and how’s the weather back home?” One utterance, four actions, zero follow-up prompts. Gurman’s sources say Apple’s test rigs already pull this off by feeding the acoustic stream into the Gemini-derived stack, then dispatching APIs across first-party apps. Privacy hawks will note that the demo keeps everything on-device until the model needs web context—Apple’s way of whispering, “We still hate cloud dossiers.”
There’s more. A standalone Siri app is in parallel development, meant to break the assistant out of Settings ghettos and give power users a prompt window that behaves like ChatGPT. Keyboards are also getting an AI rewrite: the autocorrect engine will surface synonym suggestions in-line, a move to quell thumb-fatigue and keep users inside Apple’s walled garden rather than fleeing to Grammarly.
The $1 billion question
Partnering with Google for generative brains is pragmatic, but it undercuts Apple’s years-long “privacy-first” drumbeat. Every Gemini inference still touches Google silicon somewhere, even if Apple masks IP addresses and deletes logs after 24 hours. Meanwhile, OpenAI courts Samsung and Microsoft flirts with Meta; the AI cold war is now fought over default-assistant real estate on two billion pockets.
Apple’s last-minute sprint shows how far behind it trailed. The company that once mocked cloud-reliant voice assistants is now licensing the very same tech to survive the next decade. If the 1.2-trillion-parameter Siri debuts on schedule, September’s iPhone 17 keynote could open with the first on-stage query that actually works—spoken once, executed everywhere, no apology letter needed.