o1/o3 were the real gpt5 and they did delivered hard maybe bigger jump than gpt3 to 4, RL is still continuing to follow scaling law Pretraining also scale but inference is just too expensive with giant model Agree tho than open source will win