OpenAI announced GPT-4, a large multimodal model accepting image and text inputs and emitting text outputs. It exhibited human-level performance on many professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. It launched via ChatGPT Plus and a waitlisted API. Citing the competitive landscape and safety, the technical report withheld details such as model size and training methods, a break from OpenAI's earlier openness that drew intense debate.