Apple considers starting from scratch, using external AI models to support the new version of Siri

In order to reverse the troubled AI project, Apple is considering abandoning its own AI model and using Anthropic PBC or OpenAI's AI technology to support the next generation of Siri.

People familiar with the matter revealed that Apple has discussed with the two companies to use their large language models for Siri and asked them to train a version of the model that can run on Apple's cloud platform for testing.

If an external model is eventually adopted, it will mark a major shift for Apple. Currently, most of Apple's artificial intelligence functions rely on its self-developed "Apple Foundation Models" technology, and it was originally planned to launch a new version of Siri based on this technology in 2026.

Switching to Anthropic's Claude or OpenAI's ChatGPT means that Apple admits that it is not competitive enough in the field of generative AI. Apple has allowed ChatGPT to process web search queries in Siri, but Siri itself is still driven by Apple technology.

Apple software chief Craig Federighi said at the Worldwide Developers Conference in June that it is considering whether to expand its use of external artificial intelligence technology.

People familiar with the matter said that Apple's evaluation of third-party models is still in the early stages and the company has not made a final decision on whether to adopt them. Another competing project within the company, called "LLM Siri," uses Apple's own model and is still under active development.

If the model change is implemented next year, Apple is expected to make Siri's functions comparable to Android phone AI assistants.

Representatives of Apple, Anthropic and OpenAI all declined to comment.

People familiar with the matter revealed that Apple's AI team is currently in a state of high uncertainty and lack of clear direction, and management is still deeply studying multiple possible development paths. Apple has approved a budget of billions of dollars in 2026 for running its own models through the cloud, but long-term planning is still unclear.