OpenAI launches the multimodal GPT-4
GPT-4 accepts image and text input, improves performance on complex instructions and professional exams, and becomes OpenAI's new product foundation.
What happened
Event details
GPT-4 was released as a large multimodal model that could process text and images while producing text output. OpenAI reported substantial gains over GPT-3.5 on complex instruction following and a range of professional and academic benchmarks, alongside extensive safety testing and mitigation work. The company disclosed far less about architecture and training data than earlier GPT releases, marking a broader shift toward closed frontier-model development.
Assessment
Why it matters
GPT-4 brought multimodal input and stronger general reasoning into mainstream products while making the frontier-model race more closed and capital intensive.