Apple has taken a significant step in expanding its artificial intelligence ecosystem by announcing that developers will gain access to the company's on-device foundation models. The announcement came during the Worldwide Developers Conference (WWDC) keynote on June 9, 2025, marking a strategic shift in Apple's approach to AI development.
The foundation models being opened to developers are the same ~3B parameter models that Apple currently uses internally to power features like text summarization and autocorrect in its Apple Intelligence platform. By providing developers with a software development kit (SDK) and related frameworks, Apple is enabling them to integrate these AI capabilities directly into their applications.
This move represents a meaningful milestone for Apple's AI strategy. While these on-device models are significantly smaller and less capable than cloud-based systems from competitors like OpenAI and Google, they offer distinct advantages. The models run directly on users' devices rather than in the cloud, aligning with Apple's longstanding commitment to privacy and on-device processing.
Industry analysts note that this approach could potentially unlock genuinely useful features for users while maintaining Apple's privacy-first philosophy. The on-device implementation means developers can create more intelligent applications without sending sensitive user data to external servers.
Behind the scenes, Apple has been developing more powerful models, with internal testing of versions ranging from 3B to 150B parameters. However, only the smaller on-device models are being made available to developers at this time.
While Apple continues to face criticism for lagging behind competitors in the AI race, this developer-focused initiative represents an important step in building out the company's AI ecosystem. By empowering developers to create more intelligent applications within Apple's framework, the company is laying groundwork for a more AI-capable platform while maintaining its distinctive approach to user privacy.