OpenAI announced on Wednesday that its popular ChatGPT and Whisper AI models now have developer APIs available for developers to use in their applications. A set of protocols known as an API (application programming interface) enables various computer programs to communicate with one another. For a recurring fee that is based on usage, app developers can use OpenAI technology to improve their apps’ capabilities.
ChatGPT, which was released toward the end of November, produces coherent text in a variety of styles. The speech-to-text model Whisper can convert spoken audio into text. It was released in September.
In particular, there has been a lot of demand for a ChatGPT API, which led to the creation of an unlicensed API at the end of last year that broke OpenAI’s terms of service. To meet demand, OpenAI has now launched its own API product. Off-device and in the cloud, compute for the APIs will take place.
OpenAI’s new ChatGPT API model, which replaces its previous “best” LLM API, “text-davinci-003, is referred to as “gpt-3.5-turbo.” OpenAI claims that at $0.002 per 1,000 tokens (approximately 750 words), it is approximately ten times less expensive than its existing GPT-3.5 models. “Through a series of system-wide optimizations, we’ve achieved 90% cost reduction for ChatGPT since December,” says OpenAI on its API announcement page.
Snapchat, whose “My AI” bot we talked about on Monday, is one of the early adopters of the ChatGPT API; an educational platform known as Quizlet that is said to aid students in their studies; and Instacart, which intends to include “Ask Instacart” so that customers can inquire about food later this year.
The open-source whisper-large-v2 model serves as the foundation for the Whisper API, which can be purchased for $0.006 per minute. It can translate audio into text or transcribe at a rate comparable to that of a skilled human transcriptionist, even when dealing with difficult audio, and it accepts inputs in the M4A, MP3, MP4, MPEG, MPGA, WAV, and WEBM formats.
OpenAI claims that it has amended its Terms of Service (TOS) to prohibit the use of data submitted for service enhancements, including future model training, in response to customer privacy concerns. In addition, API users are subject to a default 30-day data retention policy, and the TOS emphasizes that users own the AI models’ input and output.
“For users who want deeper control over the specific model version and system performance,” OpenAI now offers dedicated cloud instances running on Azure for its AI models.
The Whisper and ChatGPT APIs are now available. An OpenAI account and a payment method are required for use.