Questions about GPT-6 training compute and parameter estimates

A user says EpochAI released estimates for GPT-6's training compute and parameter count. They find the total compute estimate reasonable but question the 5.35x pre-training-to-RL compute figure, the 42–135-day 90% interval for pre-training, and the parameter estimates. They suggest other models may have shared Stargate Abilene's GPUs. In a follow-up, they clarify that their proposed early May-to-late July window covers both pre-training and post-training/RL.