When OpenAI published the GPT-4 Technical Report in March 2023, it deliberately withheld the model’s basic engineering facts. The report states: “Given both the competitive landscape and the safety implications of large-scale models like GPT-4, this report contains no further details about the architecture (including model size), hardware, training compute, dataset construction, training method, or similar.” The number of parameters in GPT-4, the data it was trained on, and the compute used to train it have never been officially confirmed by OpenAI.
This marked a clear departure from the field’s earlier norms. The 2020 GPT-3 paper, by contrast, openly reported its 175 billion parameters and described its training data, and academic machine-learning culture had long treated such details as the substance of a publication. The GPT-4 report’s reticence - from a company whose name still carried the word “Open” - became a frequently cited example of how commercial competition was reshaping what frontier labs were willing to publish.