--- title: Apple’s Foundation Models and Why They’re Different From... description: Apple's Foundation Models represent a fundamentally different AI strategy from OpenAI, Anthropic, and Google. Rather than competing on cloud-based model size,... canonical: https://daily.dev/posts/apple-s-foundation-models-and-why-they-re-different-from-other-ai-providers-2gck4h5yv twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Apple’s Foundation Models and Why They’re Different From Other AI Providers | daily.dev og:description: Apple's Foundation Models represent a fundamentally different AI strategy from OpenAI, Anthropic, and Google. Rather than competing on cloud-based model size,... og:url: https://daily.dev/posts/apple-s-foundation-models-and-why-they-re-different-from-other-ai-providers-2gck4h5yv og:image: https://api.daily.dev/og/posts/2Gck4h5Yv.png og:image:alt: Apple’s Foundation Models and Why They’re Different From Other AI Providers og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Apple’s Foundation Models and Why They’re Different From Other AI Providers **[Medium](https://daily.dev/sources/medium_js)** · 32 min read · 1 upvotes · 0 comments ## Summary Apple's Foundation Models represent a fundamentally different AI strategy from OpenAI, Anthropic, and Google. Rather than competing on cloud-based model size, Apple built a family of on-device models optimized for its own hardware stack — leveraging vertical integration across chip (Neural Engine), OS (iOS/macOS), developer framework (FoundationModels), and App Store. The third-generation lineup includes a 3B dense Core model for classification/tagging and a 20B sparse Core Advanced model using Instruction-Following Pruning (IFP), which routes once per prompt to activate only 1–4B parameters, achieving ~9B-class quality at 3B compute cost. The FoundationModels Swift framework offers structured generation with @Generable types, tool calling, multimodal input, and session-based conversations — all with no API key or network required. A walkthrough builds AFMChat, a minimal on-device SwiftUI chat app using XcodeGen and a Makefile-driven workflow, demonstrating the full stack running offline on iPhone 17 Pro hardware. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is the difference between Apple's Core and Core Advanced foundation models in iOS 26? Core is a 3-billion-parameter dense model reached via SystemLanguageModel(useCase: .contentTagging), built for fast tagging and classification tasks. Core Advanced is a 20-billion-parameter sparse model behind SystemLanguageModel.default, using Instruction-Following Pruning to activate only 1 to 4 billion parameters per prompt, delivering roughly 9B-class quality at 3B compute cost, and requires 12GB of unified memory. _Developers weighing on-device AI trade-offs can track how Apple's model tiers evolve on daily.dev._ ### What hardware is required to run Apple's Core Advanced foundation model on iOS 26? Core Advanced needs 12GB of unified memory, which narrows eligible devices to the iPhone 17 Pro, iPhone 17 Pro Max, iPhone Air, plus Macs with M3 or later chips and iPads with M4 or later chips. An iPhone 15 Pro or iPhone 16 can only access the smaller Core model, not Core Advanced, regardless of iOS version. _Anyone scoping device support for on-device AI features can keep hardware requirement changes like this on daily.dev._ ### What is Instruction-Following Pruning (IFP) and how does it differ from Mixture-of-Experts? Instruction-Following Pruning is Apple's sparse activation technique that analyzes a prompt once before generation begins, using a small predictor to decide which parts of the network are needed, then holds that decision fixed for the entire response. Unlike Mixture-of-Experts, which routes per token and keeps all experts resident in memory, IFP makes one routing decision upfront, reducing memory movement, energy use, and DRAM pressure on phones. _Engineers comparing sparse model architectures for memory-constrained devices can follow developments like IFP on daily.dev._ ## Similar posts on daily.dev - [Apple Launches Core AI for Apple-Silicon Optimized On-Device Generative AI](https://daily.dev/posts/apple-launches-core-ai-for-apple-silicon-optimized-on-device-generative-ai-dfllineyn) · InfoQ · 1 upvotes · 0 comments - [When to Use Apple Foundation Models on Mobile](https://daily.dev/posts/when-to-use-apple-foundation-models-on-mobile-nuyxlcjm5) · Callstack Blog · 1 upvotes · 0 comments - [Local AI Needs to be the Norm .](https://daily.dev/posts/local-ai-needs-to-be-the-norm--pjzpaddod) · Hacker News · 7 upvotes · 0 comments - [Apple bets cheaper AI will woo small developers](https://daily.dev/posts/apple-bets-cheaper-ai-will-woo-small-developers-3whrlbqpg) · TechCrunch · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#swift](https://daily.dev/tags/swift), [#apple-intelligence](https://daily.dev/tags/apple-intelligence) [View this post on daily.dev](https://daily.dev/posts/apple-s-foundation-models-and-why-they-re-different-from-other-ai-providers-2gck4h5yv) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Apple’s Foundation Models and Why They’re Different From Other AI Providers","url":"https://daily.dev/posts/apple-s-foundation-models-and-why-they-re-different-from-other-ai-providers-2gck4h5yv","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/apple-s-foundation-models-and-why-they-re-different-from-other-ai-providers-2gck4h5yv"},"datePublished":"2026-07-20T23:55:40.512Z","dateModified":"2026-09-14T08:53:16.231Z","description":"Apple's Foundation Models represent a fundamentally different AI strategy from OpenAI, Anthropic, and Google. Rather than competing on cloud-based model size,...","image":"https://media.daily.dev/image/upload/s--CxzD6vbw--/f_auto/v1722860399/public/Placeholder%2005","thumbnailUrl":"https://media.daily.dev/image/upload/s--CxzD6vbw--/f_auto/v1722860399/public/Placeholder%2005","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/apple-s-foundation-models-and-why-they-re-different-from-other-ai-providers-2gck4h5yv","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,swift,apple-intelligence","timeRequired":"PT32M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"Apple’s Foundation Models and Why They’re Different From Other AI Providers"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/apple-s-foundation-models-and-why-they-re-different-from-other-ai-providers-2gck4h5yv#faq","mainEntity":[{"@type":"Question","name":"What is the difference between Apple's Core and Core Advanced foundation models in iOS 26?","acceptedAnswer":{"@type":"Answer","text":"Core is a 3-billion-parameter dense model reached via SystemLanguageModel(useCase: .contentTagging), built for fast tagging and classification tasks. Core Advanced is a 20-billion-parameter sparse model behind SystemLanguageModel.default, using Instruction-Following Pruning to activate only 1 to 4 billion parameters per prompt, delivering roughly 9B-class quality at 3B compute cost, and requires 12GB of unified memory. Developers weighing on-device AI trade-offs can track how Apple's model tiers evolve on daily.dev."}},{"@type":"Question","name":"What hardware is required to run Apple's Core Advanced foundation model on iOS 26?","acceptedAnswer":{"@type":"Answer","text":"Core Advanced needs 12GB of unified memory, which narrows eligible devices to the iPhone 17 Pro, iPhone 17 Pro Max, iPhone Air, plus Macs with M3 or later chips and iPads with M4 or later chips. An iPhone 15 Pro or iPhone 16 can only access the smaller Core model, not Core Advanced, regardless of iOS version. Anyone scoping device support for on-device AI features can keep hardware requirement changes like this on daily.dev."}},{"@type":"Question","name":"What is Instruction-Following Pruning (IFP) and how does it differ from Mixture-of-Experts?","acceptedAnswer":{"@type":"Answer","text":"Instruction-Following Pruning is Apple's sparse activation technique that analyzes a prompt once before generation begins, using a small predictor to decide which parts of the network are needed, then holds that decision fixed for the entire response. Unlike Mixture-of-Experts, which routes per token and keeps all experts resident in memory, IFP makes one routing decision upfront, reducing memory movement, energy use, and DRAM pressure on phones. Engineers comparing sparse model architectures for memory-constrained devices can follow developments like IFP on daily.dev."}}]} ```