OpenAI launches GPT-4 with staged access to text and image capabilities
OpenAI launches GPT-4 as a model that accepts text and images and emits text; text access begins through ChatGPT Plus and an API waitlist, while image input remains in limited partner testing.
What happened
Event details
On March 14, 2023, OpenAI launched GPT-4. The company described it as a multimodal model that accepts text and images and emits text, and reported stronger results than GPT-3.5 on professional and academic exams, traditional machine-learning benchmarks, and multilingual MMLU. These results were primarily vendor-run evaluations. The launch API offered an 8,192-token model and limited access to a 32,768-token version. Availability was narrower than a general public release: ChatGPT Plus users faced usage caps, API access required a waitlist and staged invitations, and image input was still an unavailable research preview outside limited partner testing. OpenAI also acknowledged hallucinations, reasoning errors, and biased output. The technical report described a pretrained and RLHF-tuned Transformer-style model but explicitly withheld model size, detailed architecture, hardware, training compute, and data-construction details. The catalog has therefore removed the unsupported 1.8T and MoE claims.
Assessment
Why it matters
GPT-4 brought stronger general-purpose performance, multimodal input, and system-message control into ChatGPT and the developer API, accelerating product and application upgrades. Its impact was broad and durable, although launch access was constrained and technical transparency remained limited.
Availability
Access notes
ChatGPT Plus had usage caps, API access required a waitlist, image input was in limited alpha, and the 32K model had limited availability.