OpenAI opens new APIs for ChatGPT and Whisper

The new ChatGPT API calls the gpt-3.5-turbo model, which is now being used in the ChatGPT product. The company also describes it as its best model for many non-chat use cases

Basically, the system takes a series of messages as input and returns an AI model-generated message as output. More information about the ChatGPT API can be found here. An example of what an API call looks like is pictured above (using OpenAI Python v0.27).

Usage is priced at $0.002 per 1K tokens, which OpenAI says is 10x cheaper than for the existing GPT-3.5 model.

GPT stands for a “generative pre-trained transformer” family of language models. With the AI company being financially backed by Microsoft, its technology is being worked into Microsoft products such as the voice assistant and browser, Bing and Edge.

Policies

Note that in terms of its OpenAI’s user policies changing, data submitted through the API is no longer used for model training unless users explicitly opt in.

Also, any data sent through the API will be retained for a maximum of 30 days, for abuse and misuse monitoring purposes, after which it will be deleted, says the company.

You can read more here.

Whisper API

For the speech-to-text of Whisper, it says a large-v2 model is available through the API as whisper-1. This is intended to be the most convenient way to use the model. Whisper API documentation can be found here.

Note, the system only supports input files that are less than 25 MB. Audio files longer than that will need to be broken up into chunks of 25 MBs or less, or require a more compressed audio format.

In terms of pricing, use of the API is priced at $0.006 / minute, rounded up to the nearest second.

See also: OpenAI’s Evolving Ecosystem