Give up self-developed models? Apple considers GPT or Claude to support the new Siri

Apple may consider abandoning its self-developed AI models and outsourcing Siri to Anthropic or OpenAI.

According to media reports on Monday, Apple has begun negotiations with Anthropic and OpenAI to explore the possibility of integrating their language models into Siri. According to insiders, Apple has requested that both companies train customized versions of the models to run and test on Apple's own cloud infrastructure.

If Apple ultimately decides to move forward with this plan, it would represent a milestone strategic reversal. The company currently relies primarily on its self-developed technology known as Apple Foundation Models to drive AI capabilities, with plans to launch a new version of Siri based on this technology in 2026.

Following the news, Apple's stock price rose by more than 2%, reflecting investor optimism about this potential collaboration.

Internal Testing Shows Anthropic Technology is Superior

According to insiders, under the leadership of Siri team head Mike Rockwell and software engineering head Craig Federighi, Apple has initiated an evaluation project for external models. After multiple rounds of testing, Rockwell and other executives concluded that Anthropic's Claude technology best meets Siri's needs.

This evaluation result prompted Apple's Vice President of Corporate Development Adrian Perica to begin preliminary negotiations with Anthropic. Insiders indicate that Apple is also in contact with OpenAI to ensure multiple options.

Media reports suggest that executives like Federighi and Rockwell believe that instead of sticking to underperforming in-house technology, it is better to quickly enhance Siri's capabilities through collaboration.

Commercial Negotiations Have Differences, Multiple Plans Progressing Simultaneously

In terms of technical implementation, Apple has a clear plan for integrating external models.

Reports indicate that in discussions with Anthropic and OpenAI, Apple has requested that the customized versions of Claude and ChatGPT be able to run on its Private Cloud Compute servers.

These infrastructures, based on high-end Mac chips, are currently used to run Apple's more complex self-developed models. Apple believes that its own cloud servers can better protect user privacy compared to third-party infrastructures.

However, there are differences between Apple and Anthropic regarding initial financial terms. According to insiders, the AI startup is seeking fees in the billions of dollars annually, with significant increases each year. The difficulties in negotiations have led Apple to consider collaborating with OpenAI or other companies.

Currently, Apple's Apple Intelligence feature employs a dual model architecture that operates both on-device and in the cloud. The on-device model handles lightweight tasks such as email summarization and Genmoji creation, while more complex tasks rely on the cloud model Media reports indicate that Apple plans to open device-side models to developers later this year, but there are currently no similar arrangements for cloud models due to server capacity limitations.

It is worth noting that Apple has not completely abandoned its self-developed models.

According to informed sources, proprietary models will still be at the core of device-side and developer use cases, and Apple has approved a budget of billions of dollars for the operation of cloud self-developed models in 2026. However, if Siri shifts to external technologies, there are concerns within the team that this may signal similar adjustments for other functions in the future.

Siri upgrades frequently delayed, AI team morale affected

Since its launch in 2011, Siri has clearly fallen behind in the competition among AI chatbots.

Last year, the new features for Siri announced by Apple, including the ability to access user personal data and analyze screen content, were originally scheduled for release in early 2025 but have ultimately been indefinitely postponed until next spring.

Insiders from Apple's AI team stated that the team is currently in a state of high uncertainty, lacking clear direction, and executives are still weighing various possible paths. Although Apple has approved a budget of billions of dollars for cloud self-developed models in 2026, longer-term plans remain unclear.

The potential shift in Apple's AI strategy has significantly impacted the morale of the internal team.

Media reports suggest that the approximately 100-member foundational model team at Apple is dissatisfied with the company's consideration of external technologies, believing it implies inadequacies in their work.

Meanwhile, Apple's AI department is undergoing a management restructuring.

The influence of AI chief John Giannandrea continues to wane; in addition to losing responsibilities for Siri, he has also been stripped of management authority over the robotics department. The Core ML and App Intents teams have also been transferred to Federighi's software engineering department.

To make matters worse, the issue of talent loss is also intensifying.

Some members of the team have explicitly stated that if the company fully shifts to external technologies, they may accept lucrative offers from Meta or OpenAI. Recently, senior language model researcher Tom Gunter left, and the MLX open-source system team nearly left en masse, further exacerbating Apple's talent crisis