Apple Offers Free AI Access to Its AI Models for Smaller App Developers

At its annual developer conference WWDC, Apple has unveiled a new initiative designed to make it easier for smaller app developers to get started with AI development. Developers who meet certain requirements will be able to use Apple’s AI infrastructure free of charge in the future.

Apple is enabling developers enrolled in the App Store Small Business Program whose apps have so far recorded fewer than two million first-time downloads to access the next generation of Apple Foundation Models at no cost. These run on Apple’s Private Cloud Compute, meaning no cloud API costs are incurred.

The two-million threshold is based on an already familiar principle: Apple’s Small Business Program likewise grants smaller developers reduced commissions to lower the barrier to entry for app development.

Distilled Models from Google

The new Apple Foundation Models (AFM) are the third generation of Apple’s AI models — and, as is known, they are not developed in-house, but rather a suite of five base models tailored in collaboration with Google. “These range from models that run on the device itself to server-based models that run on Private Cloud Compute,” the company states.

On-Device Models:

• AFM 3 Core — Apple’s next-generation 3-billion-parameter dense model; runs entirely on-device and delivers a clear quality leap over the previous generation.
• AFM 3 Core Advanced — Apple’s most powerful on-device model, natively multimodal; a 20-billion-parameter model with a sparse architecture that activates only 1–4 billion parameters depending on the request; enables features such as expressive voices and more precise dictation, optimized for the most capable Apple Silicon chips.

Server Models (Private Cloud Compute):

• AFM 3 Cloud — the server-side “workhorse,” optimized for speed, efficiency, and performance.
• ADM 3 Cloud (Image) — Apple’s model for image generation and editing; powers advanced photo editing tools and the new Image Playground. (Note: Apple’s original text actually reads “ADM,” not “AFM” — presumably Apple’s diffusion model, not a typo.)
• AFM 3 Cloud Pro — the most powerful server model for the most demanding use cases such as agentic tool use and complex reasoning; runs, according to Apple, on Nvidia GPUs in Google Cloud and is comparable to Google’s Gemini Frontier models.

Technical Extensions to the Foundation Models Framework

In addition to free access for smaller developers, Apple has also expanded the Foundation Models Framework itself. The key new features include:

• Multimodal inputs: Developers can now pass images alongside text, enabling apps to process visual content.
• Server model support: The API can now be connected to external cloud providers such as Claude or Gemini.
• Dynamic Profiles: Models, tools, and instructions can be flexibly swapped out within a running session.
• Evaluations Framework: A new testing tool verifies whether AI features function correctly under dynamic conditions.

Context: Rising AI Costs as an Industry Issue

Apple’s initiative comes at a time when the costs of AI experimentation are facing increasing scrutiny across the industry. Companies such as Meta and Amazon have abolished internal leaderboards on which developers previously competed in their consumption of AI computing capacity. Uber recently reported having exhausted its entire AI budget for 2026 after just four months.

Against this backdrop, Apple is positioning its models as a cost-effective alternative for developers who do not want to take on additional cloud bills. At the same time, the company is strengthening the loyalty of smaller developers to its own ecosystem by offering them a low-barrier entry point into AI development.