“Citing Noam Brown, model performance scales significantly with the amount of compute applied during inference, a concept known as test-time compute.”