Rumor: OpenAI Cuts Inference Costs by 50%+ with New Optimizations

ericmitchellai · x · 2026-08-27

A report claims OpenAI has found internal optimizations that cut inference costs by more than 50% without swapping models. Engineers are also developing new chips with significant engineering leaps. These changes are expected to roll out in the coming months. The rumor also mentions a new model family 'Astra' and a massive 10T+ parameter run 'Bel'.

Original post →

More from Companies & People

Companies & People channel →