High reasoning settings on large models cost drastically more without better results

Recent technical evaluation indicates that setting frontier models to maximum reasoning modes often fails to solve basic tasks while raising compute costs up to 42 times. Organizations are increasingly finding that custom fine tuned open models with smaller parameter sizes outperform top tier commercial models on targeted tasks at a fraction of the operating cost.