OpenAI launches GPT-4 API for general use, retires previous model APIs

On 6 July 2023, OpenAI made GPT-4, its newest text-generating model, widely accessible by announcing its general availability. It has expanded access to the GPT-4 API, making it available to all paying API customers. As part of this update, older models of the completions API will be deprecated and will no longer receive further development or support.

In March, OpenAI introduced the ChatGPT API, enabling millions of developers to access it. Currently, existing API developers who have a track record of successful payments can access the GPT-4 API with an 8K context, that means it has a context length of 8,192 tokens or almost 12-13 pages of text.

Access to the GPT-4 API is currently restricted to developers with a proven track record of successful payments. However, OpenAI has outlined its intentions to grant access to new developers by the end of this month.

Additionally, the availability of the GPT-4 API will be expanded over time, with rate-limits being adjusted based on compute availability.

GPT-4 API general availability

GPT-4 represents a significant advancement over its predecessor, GPT-3.5, in terms of capabilities. It possesses the ability to generate text that includes code snippets and can accept both image and text inputs. The training data for GPT-4 comprises publicly available information, including data from web pages accessible to the public as well as licensed data obtained by OpenAI.

The company said that "Millions of developers have requested access to the GPT-4 API since March, and the range of innovative products leveraging GPT-4 is growing every day."

OpenAI has announced its plans to introduce the capability for developers to fine-tune both GPT-4 and GPT-3.5 Turbo using their own data. This functionality, which is already available for other text-generating models from OpenAI, is expected to be released later this year.

Text completions to chat completions transition

The Chat Completions API, in particular, gives developers the ability to create conversational experiences and a wide variety of completion tasks (e.g., system messages, function calling) and multi-turn conversation capabilities.

On the other hand, it is specifically built for generating responses in a conversational context and requires input in a specific format that includes message history.

By adopting a structured interface and enabling multi-turn conversations, developers have been able to create conversational experiences and accomplish tasks using the Chat Completions API. Additionally, this approach enhances security by reducing the risk of prompt injection attacks.

Deprecation of the Edit API

OpenAI has announced the deprecation of the Edits API and its associated models, such as text-davinci-edit-001 or code-davinci-edit-001. The Edits API was initially introduced as a beta version, aimed at allowing developers to generate an edited version of the prompt based on specific instructions.

OpenAI gathered valuable feedback from users of the Edits API, which helped inform the development of GPT-3.5-turbo and the Chat Completions API. These newer offerings can now be utilised for the same purpose as the Edits API, providing enhanced capabilities and improved performance.

More about its new update

OpenAI has made an announcement regarding the general availability of three APIs: GPT-3.5 Turbo, DALL-E, and Whisper. DALL-E 2 and Whisper are speech-to-text and image-generating models, respectively. These APIs are now accessible for developers to use and integrate into their applications.

Additionally, they are actively developing the capability of fine-tuning for both the GPT-4 and the GPT-3.5 Turbo. This upcoming feature will enable developers to customise and train these models to suit specific tasks.

Starting 4 January, 2024, the older completion models will no longer be available. OpenAI stated that their future focus for model and product improvements will be on the Chat Completions API. It is planning to continue refining and improving the Chat Completions API, focusing on future models and product enhancements.

Developers who are currently using these models will need to manually upgrade their integrations before the given deadline. If developers wish to continue using fine-tuned versions of the old models beyond the deadline, they will need to perform fine-tuning on top of the new base GPT-3 models.