On March 14, 2023, OpenAI released GPT-4. The GPT-4 Technical Report on arXiv describes it as “a large-scale, multimodal model which can accept image and text inputs and produce text outputs.” OpenAI’s announcement page carries the same March 14, 2023 date and detail.
The report states that GPT-4 “exhibits human-level performance on various professional and academic benchmarks,” including scoring in roughly the top 10 percent of test takers on a simulated bar exam, where the prior model had scored near the bottom 10 percent. The model is built on a transformer architecture trained to predict the next token, with post-training alignment to improve factuality and adherence to desired behavior.
A notable engineering claim is predictable scaling: OpenAI says it built infrastructure that could “accurately predict some aspects of GPT-4’s performance based on models trained with no more than 1/1,000th the compute.” This let them forecast capabilities before the full training run finished.
GPT-4 mattered as evidence that scaling and alignment could push a single model to broadly strong performance across exams and tasks, and its multimodal image input pointed toward systems that reason over more than text.