A new report from Mark Gurman at Bloomberg tells us that Apple is working with Google to test out a custom version of Gemini that could power Apple’s next-gen Siri voice assistant in the cloud.

The report says Apple is in “early discussions” and that Google is currently training a version of its Gemini AI assistant that can run on Apple’s cloud infrastructure. This is similar to the partnerships Apple has explored with Anthropic’s Claude or OpenAI’s ChatGPT. Apple is creating a next-gen, “LLM Siri” using its own models while at the same time developing an infrastructure that could see the new Siri powered by a third-party AI model. We wrote about that effort here.

According to Gurman’s report, Apple is still “several weeks away” from deciding whether to go with Apple’s own internal models for the new Siri or to employ tech from outside partners.

Apple is employing a “bake-off” approach, pitting the internally-developed next-gen Siri (code-named Linwood) against a next-gen Siri powered by external models (code-named Glenwood). The company has not yet decided which external technology is best, nor has come to financial terms with them. Supposedly, Anthropic’s Claude was the early favorite, but Anthropic demanded such a steep price that Apple started reaching out to other companies for alternatives.

This work to use Google’s Gemini is separate from the previously-reported effort to add Gemini as another option for general queries in Apple Intelligence. Today, ChatGPT is the only supported third-party AI for these sorts of general knowledge queries, but Apple has plans to add Gemini as another option. This new report, by contrast, is about Google developing Gemini-based “brains” to serve as the foundational model for the new Siri, running on Apple’s servers.