Speculation mounts OpenAI's mysterious 'new model' is a fresh pretrain, not an RL run

teortaxesTex · x · 2026-09-16

OpenAI's Will Depue claims internal hints point to a brand-new pretrain rather than an RL run — implying OpenAI may have a model that, under two weeks into pretraining, beat Astra and "solved Navier-Stokes."

TeortaxesTex disagrees, arguing it is likely just an RL run, but notes that with OpenAI's compute they could unironically pretrain a 1e27-FLOP model in two weeks. Numbers are educated guesses based on public hints — unconfirmed.

Original post →

More from Models

Models channel →