OpenAI Releases GPT-4 API Publicly, But API Completion Model Begins To Stop

GPT-4 is trained using publicly available data (photo: dock. special)

JAKARTA - OpenAI has finally released the Application Programting Interface (API) publicly from GPT-4, its newest text-producing model.

This is OpenAI's newest model, the GPT-4 is claimed to be the most capable. Since last March, the company has been asked by millions of developers to open access to the GPT-4 API.

The opening of this access, all existing API developers with payment history can access the GPT-4 API in the context of 8K.

"We plan to open access to new developers by the end of this month, and then start raising the tariff limit after that depending on the availability of computing," OpenAI said in its official blog.

GPT-4 can generate or receive text and image inputs, including codes that are an upgrade of its predecessor GPT-3.5. GPT-3.5 only receives text and works at the human level on various professional and academic benchmarks.

Like previous models, GPT-4 is trained using publicly available data, covering public web pages, as well as data licensed by OpenAI.

However, image understanding capabilities are not yet available to all OpenAI subscribers. The company is testing it with one partner, Be My Eyes as a start, as quoted by TechCrunch, Tuesday, July 11.

Going forward, OpenAI says it will provide developer capabilities to perfect GPT-4 and GPT-3.5 Turbo, one of the latest text-producing models, with their own data coming later this year.

"We are working on activating safe improvements to GPT-4 and GPT-3.5 Turbo and hope this feature will be available by the end of 2023," said OpenAI.

In addition, the company has also chosen to start stopping the Completion API model for the newer Chat Completions API model.

Launched in March, the Chat Completions API now reaches 97 percent of the use of the API GPT. This termination plan includes the termination of the model which is part of the Completions API in six months.

Starting January 4, 2024, the old model, only completing requests instead of being involved in dialogues such as chats, will be replaced with a newer version.

Developers who use some of these models will be asked to increase their integration manually. Currently, the old model will remain available but labeled as a legacy product.