Mats Hellqvist
Microsoft is a big supporter of OpenAI’s ChatGPT chatbot and has already integrated it into Bing and Edge as well as Skype. The latest update for Windows 11 also brings ChatGPT to the search box in the operating system taskbar. But that’s just the beginning: OpenAI just announced that the ChatGPT API and the Whisper API are now open to developers.
After some major optimizations, ChatGPT now costs 90% less to use than it did in December. This means that the gpt-3.5-turbo model (the base of ChatGPT) now costs $0.002 per 1,000 tokens. In other words, you get 500,000 tokens for a dollar.
ChatGPT API works in 99 languages
The new price makes it much cheaper for companies to build their own chatbot. But the ChatGPT API works with text. If your own bot should also react to voice commands, Whisper comes into play. Whisper turns speech into text at a cost of $0.006 per minute, or $1 for 2 hours 45 minutes. In addition, Whisper not only understands 99 languages, but can also provide an English translation. In contrast, the competition in the form of Siri and Google Assistant can only process a few languages.
One of the first customers is Snapchat, which rolled out My AI for Snapchat+ this week. This bot can provide recommendations, such as a recipe for dinner or plan a hike. You can even ask My AI to write a haiku. Instacart has also launched a ChatGPT API-based bot that suggests recipes and automatically creates the matching shopping list. Similarly, Shopify’s Shop app can help users find matching products and brands, much like a personal shopping assistant.
The ChatGPT API is also used in educational apps. Quizlet already introduced Q-Chat, a “fully adaptive AI tutor”. Q-Chat is much more flexible than fixed quiz questions in textbooks, which have to cater to all students and their needs. Speak is the fastest growing English app in South Korea and also has integrated Whisper.
Under the source links you can see the prices for the ChatGPT API. Dedicated instances are also available for businesses. Otherwise, the model runs on a shared instance whose performance varies depending on server load.
