Google DeepMind is reportedly considering a deeper restructuring that could eliminate about one-third or more of its workforce, with the cuts aimed at roles seen as redundant or poorly aligned with actual work needs. The report said the unit, which currently has about 7,000 to 8,000 employees, is also rethinking its model strategy. Google is said to have no near-term plan to update its Pro model and to scale back investment in ultra-large flagship models on the level of Fable and Opus, redirecting more resources toward Flash, a lower-cost model line that can be iterated more quickly. The shift may reflect internal resource constraints. Google DeepMind reportedly scored 0.5 out of 1 on its overall OKR (objectives and key results) in its most recent review year and has struggled to secure more large-model training capacity. Demand for TPU (Google's AI training chips) resources from core Google businesses including Search, YouTube and Gmail is also said to be heavy. The report added an internal detail that may help explain criticism of the Gemini user experience: the core Google DeepMind team had not previously treated Gemini as its own primary day-to-day model, while non-core teams were still required to use it.