IA · 15 September 2026 · 5 min read
Apple launches iOS 27 and macOS Golden Gate: the new Siri AI redefines user experience
In brief: Apple has officially released iOS 27 and macOS 27 Golden Gate, delivering a foundational overhaul of Siri powered by next-generation foundation models. Backed by the hybrid AFM 3 architecture, the update introduces deep on-screen awareness, prompt-generated automations, and direct app interactions across both desktop and mobile platforms.
by Team Mocchi's
The launch of iOS 27 and the reboot of Apple's assistant
Marking its annual major software cycle, Apple has officially released iOS 27, macOS 27 Golden Gate, watchOS 27, visionOS 27, and tvOS 27. As reported by Ars Technica, the defining highlight across the ecosystem is the long-anticipated Siri AI, a complete ground-up reconstruction of the company's voice and text assistant powered by modern large language models and personal contextual awareness.
After years of remaining primarily a voice interface for straightforward utility commands like timers or alarms, Siri is taking a noticeable leap forward. In real-world testing, as noted by TechCrunch, the overhauled assistant successfully resolves multistep queries, synthesizes context across messages and emails, and reasons about what is currently visible on the user's screen without demanding friction-heavy back-and-forth app switching.
Alongside system-wide vocal activation, Apple has also introduced a dedicated Siri application. Featuring a chat-based layout similar to mainstream standalone AI tools, it caters to users who prefer direct textual interaction while preserving full access to on-device data.
The AFM 3 architecture: local inference meets cloud fallback
Underpinning Siri AI and the broader Apple Intelligence feature set is Apple's proprietary AFM 3 (Apple Foundation Models) suite, engineered as a multi-tier pipeline to balance low latency, strict data privacy, and computing overhead:
- AFM 3 Core (on-device): a compact 3-billion-parameter model designed to handle immediate local tasks directly on the processor;
- AFM 3 Core Advanced (on-device): a 20-billion-parameter sparse model that dynamically activates between 1 and 4 billion parameters based on request complexity, preserving memory bandwidth and battery endurance;
- AFM 3 Cloud and AFM 3 Cloud Pro: high-parameter models hosted on Apple's Private Cloud Compute infrastructure, handling complex multistep reasoning and broad general-knowledge requests;
- ADM 3 Cloud (Image): a specialized cloud model powering expanded image generation and photo editing capabilities.
This tiered design ensures that sensitive personal context remains strictly within the device whenever possible, dispatching encrypted requests to remote clusters only when computational demands exceed local silicon thresholds.
On-screen context, prompt automations, and Liquid Glass refinements
A primary technical evolution is Siri's direct interplay with application interfaces. The assistant can now parse active screen contents, summarize documents, extract structured action points, and automatically assemble Shortcuts workflows or Safari extensions from natural-language prompts. For applications integrating the relevant system APIs, Siri can execute end-to-end tasks—such as tracking down an order confirmation in messaging threads and filing a return request—in a single flow.
Beyond artificial intelligence, the operating systems bring notable UI refinements. iOS and macOS include an opacity adjustment slider for the Liquid Glass aesthetic, addressing user feedback gathered over the past cycle. Furthermore, macOS 27 Golden Gate delivers native HDR support across all system UI elements, expanded external display configuration options, and a notable platform milestone: it represents the final version of macOS supporting the Rosetta translation layer for legacy Intel applications.
Mocchi's take
The production debut of Siri AI marks a fundamental evolution for software engineering: the operating system UI is shifting decisively from purely visual navigation to system-wide ambient orchestration. For businesses and software teams, integrating deeply with Apple's App Intents and context frameworks is no longer optional; products that fail to expose their features to system-level agents risk becoming invisible in users' daily workflows. Furthermore, Apple's sparse, hybrid architecture proves that practical, day-to-day AI does not require monolithic cloud models for every task. Pairing compact, specialized local models with secured remote fallbacks remains the most robust architecture for custom enterprise solutions as well.