Apple Offers Glimpse Into Its AI Model Training With New Technical Report
Apple has published a technical document outlining the inner workings of its next-generation AI models, introduced during WWDC 2025. Titled “Apple Intelligence Foundation Language Models – Tech Report 2025,” it delves into how the models were constructed, trained, and evaluated, marking a rare level of transparency from the company on its artificial intelligence efforts.
The report sheds light on a wide range of technical components, including model architecture, training data, efficiency upgrades and multilingual capabilities.
Faster On-Device AI That Uses Less Memory
Apple's revamped on-device model features around three billion parameters. This model has been split into two segments to improve performance and memory efficiency.
As noted in the document, Block 1 contained 62.5% of the total transformer layers, while Block 2 contained the remaining 37.5% of the transformer layers, but had the key and value projections removed. This translates into a 37.5% reduction in memory usage for caching, and speeds up the generation of the first token of a response. Apple claims this optimisation was achieved without compromising the model's accuracy or quality, as reported by 9to5Mac.
Cloud Model Built For Speed
Apple's server-side model, built for complex tasks using its Private Cloud Compute system, uses a unique design called Parallel-Track Mixture-of-Experts (PT-MoE). Unlike traditional models, it activates only the expert units needed for a specific task. For example, a prompt about cooking triggers only cooking-related experts, according to 9to5Mac.
Instead of a single processing path, the model runs across multiple parallel tracks. Each track contains both standard layers and special “expert” layers that switch on only when required. This makes the model faster and easier to scale.
To further improve performance, Apple added a method called Interleaving Global and Local Attention Layers, which helps the model understand both detailed and big-picture context. According to 9to5Mac, this design improves efficiency while keeping reasoning strong.
Stronger Support For Global Languages
Apple has increased its multilingual training data from 8% to 30% and expanded its vocabulary from “100k to 150k” tokens. This led to much better performance in non-English languages. To make responses more natural, Apple tested the model using prompts written by native speakers instead of translations. These changes should help tools like Apple's Writing Tools work better in more languages.
Ethical Data Collection
The report also details how Apple collected data for training its AI models, with a focus on privacy. Most of the data came from publicly available web pages, using Applebot, which respects robots.txt rules so sites can block access. Some content was also licenced, with reports suggesting deals with publishers like Conde Nast, NBC News and IAC.
For visual data, Apple gathered over 10 billion image-caption pairs, including photos, screenshots with text and handwritten notes, and improved them using its own AI tools.
-
Extracting Value From Battery Scrap Could Be Challenging For India | The Reason Why
A discarded lithium-ion battery still contains valuable materials such as lithium, nickel and cobalt. These can be recovered and reused. When such batteries are crushed and shredded, they produce a dark powder called black mass.
-
One El Nino, India Can Handle. Two Could Break The Granary
India's grain stocks can absorb one bad year, but a second El Nino could drain the buffer in two to three years, economists say.
-
India's Troika Of Good Luck | Joining The Dots
As Jensen Huang's exit drags Nvidia down, a cooling AI trade and resilient earnings have revived investor interest in India. A likely RBI rate hike looms.
-
Talking Points This Week: Leadership Shock At Nvidia, Earnings Strength, And The Fog Over Hormuz
The Anthropic IPO pricing and first trading day, Nvidia's upcoming earnings, and any fresh signals from Iran will set the tone for the following week.
-
Signature Play: The Classic Cocktail Is Dead. Or Is It?
Classic cocktails are disappearing from India's cocktail menus. But they're still being ordered, taught to bartenders and borrowed from. So, who's really having the last drink?
-
A New Samvat Wishlist: Inclusive Prosperity And Human Development
India has strengthened its economic foundations through infrastructure, digital access and enterprise. Its next growth chapter must deepen productivity, create better livelihoods and widen participation in prosperity.
-
Love, Sex Aur Dhokha: The Audacious Plot To Spy On Jensen Huang
A Taipei dinner, a Singapore chip buyer and a suspected Chinese asset: inside the alleged spy operation that got close to NVIDIA's Jensen Huang.
-
Ten Things I Learnt This Samvat
Patience can be a trap. Cheap stocks can stay cheap. Leverage can wipe you out. What did Samvat 2082 teach us? Niraj Shah writes on 10 lessons from the year gone by.
-
Talking Points This Week: Tech Nerves, FII Buying, And A Faraway Banking Shock
The global markets are not shining this Diwali. The exception in the large markets is, India.
-
Midterm Rally Lasted A Day. The Hard Reality Of A Divided US May Last Two Years | Opinion
Between a contested AI boom, a debt ceiling that arrives in 2027 and a president with every incentive to act alone, the next two years may prove more volatile than the first day's rally suggests.