Siri AI: 5 foundation models
Apple's new Siri AI is powered by 5 distinct foundation models — 2 on-device and 3 private cloud compute — ranging from 3 billion to 20 billion parameters.
Snapshot · Waveform: The MKBHD Podcast
WWDC 2026: Is Siri Actually Good Now?
Explore episode
Jun 12, 2026
Where this was said
How Apple Trains Siri: Five Models and the Gemini Deal
At 1:22:15 · chapter starts 1:19:40
David breaks down the architecture behind Siri AI's five foundation models: AFM 3 Core (3B parameters, handles simple on-device tasks like timers), AFM 3 Core Advanced (20B parameters using mixture-of-experts, restricted to iPhone Pro and Air with 12GB RAM), plus three private cloud compute models for heavier workloads including AFM 3 Cloud Pro, which specifically uses Google Cloud infrastructure and NVIDIA GPUs. The most surprising detail: Apple is not deploying Gemini directly. Instead, it licenses a 1.2 trillion parameter Gemini model for approximately $1 billion per year to 'teach' its own foundation models — a distillation strategy. Marques unpacks the analogy: if a car company hired a rival to teach them how to build cars using their own materials, the resulting vehicle would walk and talk like the teacher's car but be technically proprietary. Meanwhile, Google pays Apple ~$19 billion per year to be the default search engine, creating an extraordinary mutual financial dependency at the heart of the AI race.
Apple Is Paying Google $1B to Teach Siri
Apple isn't using Gemini directly in Siri. Instead, it's paying Google roughly $1 billion per year to license a 1.2 trillion parameter Gemini model as a teacher to distill knowledge into Apple's own foundation models. The models Apple ships are technically Apple's own — but they learned from Gemini.
Apple pays ~$1B/year to license Gemini
Apple is reportedly licensing a 1.2 trillion parameter Gemini model from Google for approximately $1 billion per year to use as a teacher for its own Apple Foundation Models.
Apple pays Google $19B/yr as default search engine
Apple reportedly receives approximately $19 billion per year from Google to be the default search engine on iPhones, while simultaneously paying Google ~$1 billion per year for Gemini model access.
Siri English-only at launch, not in EU/China on iPhone
The new Siri AI launches in English only and is unavailable on iPhone in the EU or China due to DMA regulations, though Mac users in the EU can access it.
Similar snapshots
Ad monetization for games
Ad-based monetization works well for game apps where users spend extended time in-session, as seen with Grid and Wordle.
Ads don't work for tool apps
Tool-focused apps like PuffCount are poor candidates for ad monetization because users don't stay in-session long enough.
Hard paywall is unskippable
A hard paywall is a screen that blocks all app features unless the user pays or starts a free trial — it cannot be dismissed.
Two core monetization models
Mobile apps are primarily monetized through either ads (best for games) or in-app purchases/subscriptions (best for tools).
YouTube beats every other platform
According to the episode, YouTube outperforms every other social platform for building trust and driving SaaS conversions.
Most users came from YouTube
Vasco stated that the majority of his app's user base came directly from his YouTube channel.
Boost My Domain Rating button
SEO Bot features a 'Boost My Domain Rating' button that routes users directly to Listing Bot, an example of in-product cross-selling.
All products connected to AI
The founder's entire product portfolio is AI-related, making it easier to package products attractively for directories.
AI coding viability debate as hook
The founder attached their SaaS demo to the trending debate about whether AI coding is actually good enough to build a full SaaS product.
