Chinchilla Optimal

The Chinchilla result is DeepMind's 2022 paper showing that for a fixed compute budget, model parameters and training data should grow roughly in balance. It corrected an earlier reading of Scaling Laws that emphasized parameter count and showed many models had been data-starved. The finding explains why later releases like Llama 3 and Mistral are trained on far more tokens than peers of similar size. Chinchilla-optimal scaling is now the reference point for planning frontier-model compute budgets.