According to @gdb, GPT-4.5 is state-of-the-art on a forecasting benchmark and, at current trends, models are about one year away from matching the performance of superforecasters. According to @gdb, this assessment references benchmarking shared by @Research_FRI on X.