All articles tagged with #on device ai
The Information reports Apple is exploring PrismML’s tech to shrink a 27‑billion-parameter open-source model (Qwen 3.6 from Alibaba) to run directly on an iPhone 17 Pro, enabling on-device AI without servers. PrismML plans to release an open-source model mid-July, and Apple has reportedly held talks with PrismML about using its technology as part of its ongoing, secretive AI efforts.
PrismML says it has compressed Alibaba's Qwen 3.6 into a 27‑billion‑parameter model that runs directly on an iPhone 17 Pro, reportedly the largest on-device AI model to date, highlighting a push toward on-device AI to cut cloud costs and boost privacy, though full technical details are behind a paywall.
Apple’s iOS 27 beta 3 introduces a major Siri AI upgrade powered by Apple Intelligence and Foundation Models, delivering richer conversations, some on-device AI, a Dynamic Island–integrated Siri interface, and broader app integration across the system; however, many features require newer hardware. The release also brings UI refinements (Liquid Glass), performance improvements, and tweaks to Photos, Camera, Messages, and AirPods.
Apple’s Doug Brooks says the Mac mini (and Mac Studio) are in “incredible demand” for running AI agents locally, as developers favor systems that are always on and under user control. He frames agentic AI as a whole‑chip challenge rather than just a GPU one, powered by Neural Engine and other neural accelerators to enable on‑device AI. Brooks envisions a hybrid future where some workloads run locally while others are offloaded to the cloud for privacy and cost reasons, and cites “transparent AI” features on iPhone/iPad and Mac (like Draw Things and SwingVision) as examples of AI quietly working in the background. The interview underscores Apple’s chip‑and‑software approach to push on‑device AI forward, a trend likely to accelerate after WWDC 2026.
Bloomberg’s Mark Gurman says Apple plans two Mac Studio refreshes: an M5 Ultra this year and a more substantial M7 Ultra in 2028, skipping the M6 entirely. The M7 Ultra is expected to feature a redesigned inner architecture and a better heat sink for improved thermals, while the M5 Ultra is not anticipated to have a major redesign. Memory options have been tested up to 768GB, though supply constraints could limit offerings, and launch timing for the M5 remains unclear after delays tied to memory chips and price increases.
Samsung announced UFS 5.0 storage with up to 10.8 GB/s read and 9.5 GB/s write, plus 40% better power efficiency and a smaller package, aimed at accelerating on‑device AI and overall performance. The Galaxy S27 Ultra is rumored to pair this with a 2nm chip, which could make it the first handset to adopt the standard, delivering a snappier experience. Mass production of UFS 5.0 is slated for Q4 2026, signaling a potential competitive edge for the S27 Ultra.
Apple's iOS 27 developer beta introduces an on-device AI dictation system (AFM 3 Core Advanced) with higher accuracy and improved punctuation/casing, but it remains off by default in beta 1 and is limited to newer devices: iPhone 17 Pro/Pro Max, iPhone Air, Vision Pro with M5, iPads with M4+ 12GB RAM, and Macs with M3+ 12GB RAM; the standard iPhone 17 (8GB RAM) is not supported, and it’s unclear if it will be auto-enabled in a future release.
Samsung Electronics announced the industry’s fastest UFS 5.0 storage solution, delivering up to 10.8 GB/s sequential read and 9.5 GB/s write, with more than 40% power efficiency gains versus UFS 4.1, in a compact 7.5 × 13 × 0.9 mm package (16.7% smaller). Mass production starts in Q4 in capacities up to 1 TB, enabling faster on-device AI for next-gen mobile, wearables and XR devices.
OnePlus has released a new Indian-market edition of the OnePlus 15R with 16GB RAM and 512GB storage, joining the existing 12GB RAM variants; remaining specs include a 7,400 mAh battery and Snapdragon 8 Gen 5. Priced at INR 61,999 (~$656), the new version appears to be India-exclusive for now with no confirmed US rollout, and the extra RAM could improve AI workloads and overall responsiveness.
Apple’s major Siri AI upgrade arrives with iOS 27, macOS Golden Gate, and watchOS 27 but will be limited to newer hardware. The broad rollout includes iPhone, iPad, Mac, Apple Watch, and Vision Pro, with full features (like customizable voices and the strongest on-device models) reserved for the newest devices (e.g., iPhone 17 Pro/Pro Max, iPad models with M4+ and 12GB+ RAM, Mac with M3+ and 12GB+, Vision Pro M5). Affected devices span iPhone 15–17 series, iPad with Apple silicon, Macs with Apple silicon, Apple Watch Series 9+/Ultra 2+/SE 3, and Vision Pro, with the public release expected this fall.
Apple says the most powerful on-device Siri AI features in iOS 27 require 12GB of unified memory, limiting the new expressiveness and improved dictation to iPhone 17 Pro models (and select iPads/Macs); base iPhone 17 gets the newer Siri interface but with older voices and less accurate dictation, while other AI features run across a wider range of devices with Apple Intelligence.
Apple unveils the third generation of its Foundation Models (AFM 3), spanning on-device (AFM 3 Core and AFM 3 Core Advanced) and server-based options (AFM 3 Cloud, ADM 3 Cloud, AFM 3 Cloud Pro), all designed with privacy at the core and deeply integrated into Apple operating systems. A key advance is AFM 3 Core Advanced’s sparsely activated architecture (IFP), which runs most weights from flash and loads small expert subsets into DRAM to deliver multimodal capabilities on Apple silicon. Server models leverage Private Cloud Compute (with NVIDIA GPUs for Cloud Pro) to boost speed, reasoning, and image generation. Apple emphasizes responsible AI, extensive evaluations showing strong gains over prior generations, and states that user data is not used for training. A technical report with more details is planned for later this summer.
Apple unveiled its third-gen Apple Foundation Models (AFM), with on-device cores and cloud services built on Apple Silicon; Apple says none of Google's Gemini components or infrastructure are used, and the suite includes AFM Cloud Pro for advanced tool use, running Nvidia GPUs in Google's cloud under a privacy-focused System Orchestrator that can be independently verified.
Apple unveiled iOS 27 and a new Siri AI at WWDC 2026, but its most powerful on-device AI model will only run on newer hardware: iPhone 17 Pro or iPhone Air, plus iPad with M4+ 12GB RAM or Mac with M3+ 12GB RAM. The base iPhone 17 with 8GB RAM cannot run the top model on-device; other devices can still use many features via Private Cloud Compute, though at slower speeds.
Apple unveils iOS 27 with major Photos upgrades: Clean Up becomes more capable thanks to smarter AI, and two new tools—Extend (generates content around image borders) and Reframe (adjusts angle/zoom). Edits run on a mix of on-device and cloud processing, with cloud work involved when saving Extended or Reframed results. Availability for iPhone, Mac, and Vision Pro updates is expected later this year.