Google DeepMind is considering cutting more than one-third of its workforce: pausing Pro updates in the short term, shifting resources to Flash

According to Dynamic Beating monitoring, Google DeepMind is in the process of further restructuring, potentially laying off a third or even more employees. GDM currently has around 7,000 to 8,000 people, and the restructuring will focus on streamlining positions and removing redundant staff whose actual work does not align.

The most significant change is in the model roadmap. Google does not have short-term plans to update the Pro model and will also reduce investment in flagship models at the level of Fable and Opus, shifting more resources to Flash, which is lower cost and has a faster iteration.

This may be related to the resources available to GDM. GDM has only achieved an overall OKR of 0.5/1 in the most recent evaluation year, making it challenging to obtain more resources for training larger models. Google's core businesses such as Search, YouTube, and Gmail also require a large number of TPUs.

There is also an internal detail that may explain why the Google Gemini model has a poor user experience: the GDM core team had never truly treated Gemini as the daily flagship model, instead, non-core teams were still required to use Gemini.