Updated October 24, 2026. This is a corrected historical explanation of the 2023 GPT-4 release; the original publication date is unchanged.
Current availability: OpenAI’s deprecation schedule lists the original gpt-4/gpt-4-0613 API models as shut down on October 23, 2026. This is not a statement that every later GPT-4-family model has the same status. Consult the current model catalog for a new project.
Artificial Intelligence (AI) has made significant strides in recent years, and one of the most noteworthy advancements is OpenAI’s Generative Pre-trained Transformer 4 (GPT-4). This article explains GPT-4 and its API in the context of the 2023 release. GPT-4 was the model name; “Original OpenAI GPT-4.0 API” was not a separate official product.
What Was the GPT-4 API?
The existing graphics illustrate GPT-4; they are not screenshots of a product interface.
OpenAI’s API provided developers with access to the GPT-4 language model. Put simply, an API is a set of rules and protocols that allow applications and platforms to interact with each other. In this case, OpenAI’s API enables developers to integrate the Natural Language Processing (NLP) capabilities of GPT-4 into their own applications and services.
OpenAI described GPT-4 as a multimodal model accepting text and images and producing text. However, API input support depended on the model version and release: the original GPT-4 API should not be treated as interchangeable with later vision-enabled models. The 2023 research announcement distinguished the model’s research capabilities from staged product access.
OpenAI’s technical report did not disclose GPT-4’s parameter count or detailed architecture. The previously reported 1.76-trillion figure was an unverified estimate, not an official specification. Parameter count alone also does not establish how well a model performs on a particular task.
Improvements and Abilities of GPT-4
One of the notable improvements of GPT-4 over its predecessor, GPT-3.5, was improved performance on many evaluated tasks. External actions were a separate integration concern: compatible models could propose function arguments, but the application had to execute the tool or API call and return its result. Browsing required a connected retrieval tool, and image generation required a separate image system; these were not automatic outputs of the base GPT-4 text model. OpenAI’s June 2023 function-calling announcement covered both GPT-4 and GPT-3.5 Turbo.
A system message lets developers specify instructions such as the model’s tone and task. It was not a capability exclusive to GPT-4; OpenAI also documented system-message steerability for GPT-3.5 Turbo. For example, a developer could use a system message to instruct GPT-4 to respond to user queries in a formal tone or to generate a summary of a news article.
Applications of GPT-4
GPT-4 supported applications such as chatbots, drafting and summarization. A developer could combine a compatible model with retrieval or other tools to build a virtual assistant, while validating sources, permissions and outputs. A model response alone did not prove that an external action had occurred.
Furthermore, GPT-4 can be helpful in the field of programming. It can assist developers in coding tasks and suggest optimizations, which can save significant time and effort.
Conclusion
GPT-4 was an important 2023 model release, but historical capabilities should not be confused with current API availability. For a new integration, check the exact supported model identifier, input and output types, tool-execution responsibilities and current pricing. Test a small representative task before relying on the result.
