OpenAI Launches Low-Cost APIs for ChatGPT, Whisper Access
New APIs bring a promise not to use your data for model training — unless you expressly permit it.
OpenAI has announced the launch of application programming interfaces (APIs) for its popular ChatGPT large language model system and Whisper voice-to-text transcription engine — claiming to have cut operational costs by 90 percent since launch.
"ChatGPT and Whisper models are now available on our API, giving developers access to cutting-edge language (not just chat!) and speech-to-text capabilities," OpenAI says in its announcement. "Through a series of system-wide optimizations, we’ve achieved 90 percent cost reduction for ChatGPT since December; we’re now passing through those savings to API users."
ChatGPT, based on the company's GPT-3.5 large language model, has been getting considerable interest of late for its ability to handle complex conversational topics — as well as its tendency to spout nonsense in a very confident manner, making facts up on-the-spot in response to user questions. While several large companies, including Microsoft, have been experimenting with integrating ChatGPT technology into their own creations, it's now generally available — at a cost of $0.002 per 1,000 tokens, ten times less than OpenAI's earlier GPT-3.5 offerings.
Even before the launch of an official ChatGPT API, the model and its predecessors had been used in a range of projects — from putting the chat-bot on amateur radio with a text-to-speech engine to driving a typewriter which could invent stories for you on-the-spot. It was also at the heart of recent work at Microsoft to make robot and drone control more accessible through natural-language instruction prompting, which it says shows considerable promise for the future.
Whisper, meanwhile, was released under an open-source license by OpenAI late last year, promising automatic speech recognition with improved accuracy and better resistance to background noise than its competitors. For those who don't want to host it locally, the new Whisper API uses the whisper-large-v2 model and offers on-demand access at $0.006 per minute — complete with the ability to transcribe in an audio file's source language or automatically translate it to English.
In an effort to boost adoption of its APIs, OpenAI has also made a new pledge: it will no longer train its models on data submitted by API users, unless the user explicitly opts-in to such use. Data is also subject to a 30-day retention policy, the company has confirmed, and stability of the service has been improved with less downtime expected in the future.
More information on the new APIs, which come with Python bindings, is available on the OpenAI website.
