FundaAI: GPT-6 training compute estimated to be 10 times that of GPT-5

Pequity Research, citing FundaAI, reports that the training compute for GPT-6 is approximately 10 times that of GPT-5. FundaAI notes that based on industry discussions, compute consumed by experiments, reinforcement learning, synthetic data generation, and supporting infrastructure could be several times higher than the main pre-training run, potentially reaching up to ten times. The next phase of pre-training acceleration is expected to begin once TPU 8t and Vera Rubin chips are shipped at scale.