GPT-4 API Officially Launched, Support for Legacy Models in the Completions API Ends

The GPT-4 API has been officially launched for all paying customers. Developers with a history of payments can now use the GPT-4 API, which supports an 8K context, immediately, and access will be expanded to new developers starting later this month. In addition, the GPT-3.5 Turbo, DALL·E, and Whisper APIs have also been officially launched, and fine-tuning capabilities for GPT-4 and GPT-3.5 Turbo are expected to be available by the end of this year.

OpenAI is accelerating the transition from the existing free-form text-based Completions API to the Chat Completions API, which provides a structured interface. The Chat Completions API currently accounts for 97% of total API usage, offering greater flexibility and security through system messages and function calling capabilities.

As part of this strategy, legacy models in the existing Completions API will have their support phased out step by step. Going forward, all model improvements will be focused on the Chat Completions API, and the main model replacement plans are as follows:

• ada, babbage $\rightarrow$ babbage-002
• curie, davinci $\rightarrow$ davinci-002
• text-davinci series $\rightarrow$ gpt-3.5-turbo-instruct

Applications using stable models such as ada, babbage, curie, and davinci will be automatically upgraded to the new models, but developers using models such as text-davinci-003 will need to manually modify the model parameter to upgrade.