OpenAI releases GPT-4 with text and image input capabilities

OpenAI released GPT-4 on 14 March 2023, making the model available to ChatGPT Plus users and selected developers through an application programming interface. It could accept text and image inputs and produce text output, although image-input access was initially limited.

The company reported stronger performance than earlier systems on many professional and academic benchmarks. Benchmarks measure defined tasks and do not establish general understanding, consciousness or reliable real-world judgment. GPT-4 could still hallucinate facts, make reasoning errors and reproduce biases.

OpenAI published a technical report and a system card describing evaluation and safety work but withheld some architecture, dataset and compute details. That limited independent reproducibility and became part of debate about transparency. Release claims should therefore distinguish company evaluations from later external research.

The model was not identical to the ChatGPT product, which could route users to different models and add tools over time. Later versions, larger context windows and multimodal products should not be treated as capabilities present in the original release.

This record captures the introduction of GPT-4 as a model and service milestone. Nothing here establishes that one launch caused every subsequent investment or policy response. Measured adoption and downstream effects require separate evidence, while the official materials establish availability, tested capabilities and disclosed limitations on the day.