Reports indicate that Google's DeepMind division is stepping back from developing frontier AI models, instead prioritizing the more cost-effective Flash-level models. Despite the Silicon Valley tradition of "eating your own dog food," the core DeepMind team has never truly adopted Gemini as their primary model for daily use. Non-core teams, however, are still required to use Gemini.

As part of the restructuring, DeepMind could face significant layoffs, potentially affecting one-third or more of its workforce. Insiders suggest the reorganization aims to eliminate redundant roles, such as employees hired for algorithm positions who are not actually performing algorithm-related work.

Alphabet, the parent company of Google, announced on August 5th a restructuring of its AI subsidiary, DeepMind. CEO Demis Hassabis has stepped down from day-to-day management to become Chairman and Chief Scientist at Alphabet, focusing on AGI strategy and scientific applications. According to reports, the restructuring is linked to delays in the development of Google's next-generation flagship Gemini model. Sources say the Gemini team has experienced internal conflicts, with multiple project leaders disagreeing on development priorities and resource allocation. Additionally, Google faces constraints in computing power for AI model training.