Open AI made the launch official of an API that will allow developers and companies to integrate ChatGPT into their applications and services. After the explosion in popularity, the company is looking to capitalize and one of the best strategies is for third parties to access its AI. For a modest fee, those interested may use the GPT-3.5-Turbo modelthe same one that drives the chatbot.
Although the fashion is to mount ChatGPT wherever possible, OpenAI says that its API can be used for more than just a chat interface. The company showed some examples of what some companies that already have access to the AI model are doing. Among them is Snap, who a few days ago announced My AI for Snapchat+, a customizable chatbot that offers fun recommendations.
API early adopters play it safe and propose experiences similar to what we see with ChatGPT. Quizlet offers a chatbot in the form of an AI teacher, who asks students questions based on the materials consulted. Instacart allows users to ask about food and get recommendations for products and recipes. While Shop, the Shopify app uses the API to power a personalized shopping assistant
The ChatGPT API will have a cost of $0.002 per 1,000 tokenswhich according to OpenAI, is ten times cheaper than existing GPT-3.5 models. GPT models process text using tokens, which are sequences of characters. A token corresponds to approximately 4 characters of text for the English language, so 1,000 tokens would equal about 750 words.
The new API will open the doors to thousands of apps with ChatGPT
OpenAI said that developers using GPT-3.5-Turbo will always end up with a stable model recommended. Those looking for more control will have dedicated instances at their disposal. This feature provides greater infrastructure for apps that process millions of tokensreducing the response time to requests.
Greg Brockman, president of OpenAI, said in a chat with TechCrunch that offering an API was always in the company’s plans. GPT-3.5-Turbo is a model optimized to respond faster. According to Kevin Menear, a generative AI experta query that normally takes 1 minute for ChatGPT, would take 16 seconds for the Turbo model.
Although ChatGPT is the most important announcement of OpenAI, the company also confirmed the deployment of its API for Whisper, a speech-to-text model capable of transcribing from audio files. For a cost of $0.006 per minute, developers will access an API that transcribes in the source language or translates into English in a variety of formats.