The costs of training and running a large language model (LLM) were prohibitively expensive, and only a few big tech companies could afford to do it.