This week, OpenAI released the gpt-3.5-turbo API, which provides access to the same model used by the ChatGPT platform. This new API is now available to developers and offers a more affordable alternative compared to other model APIs previously released by OpenAI, such as Ada, which is a faster LLM model, and Davice, which is a more powerful LLM model.
For developers and companies using OpenAI APIs to develop solutions and embed the capabilities of GPT3.5 models, such as Hop AI, the news of more affordable prices is fantastic. Previously, the costs of the APIs available were a major concern for Hop AI. Hop AI’s CEO, Hugo Azevedo, always warned clients about the issue of OpenAI API prices, recognizing that the costs could represent a significant obstacle to their adoption.
And how much was the reduction?
The new gpt-3.5-turbo API is impressively 10 times cheaper than the others previously available from OpenAI, with a price of $0.002 per 1,000 tokens. This reduction was possible due to optimizations made by OpenAI across the system since December, resulting in up to a 90% reduction in platform costs. This reduction in costs has now been passed on to set the price of this new API.
According to OpenAI, this API now accesses the best model for developing solutions for various use cases, beyond just chat use. Another piece of good news is that they claim that tests to migrate from the text-davinnci-003 API to gpt-3.5-turbo require only minor prompt adjustments. Hop AI is evaluating the new API to confirm this claim.
Operation of the gpt-3.5-turbo API
The gpt-3.5-turbo API provided by OpenAI is a tool that processes messages instead of just unstructured text, like ChatGPT and other GPT models. In other words, instead of processing only sequences of words and phrases, the model accessed by the API processes entire messages along with metadata. However, behind the scenes, the API still converts messages into token sequences so that they can be processed by the model, using a new markup language called Chat Markup Language (“ChatML”).
Chat Markup Language (“ChatML”) is a new markup language that was created specifically for use by ChatGPT. It is a way to represent the messages sent by the gpt-3.5-turbo API to the GPT-3.5 model in a way that is easily understood by the model. ChatML is built on top of HTML (HyperText Markup Language), which is a markup language used to create web pages. This means that ChatML uses a syntax similar to HTML to represent the messages sent to ChatGPT.
OpenAI is listening to the market
OpenAI also listened to developer feedback and, in addition to more attractive prices for the gpt-3.5-turbo API, it refined its API terms of service to better meet their needs.
In summary, the new gpt-3.5-turbo API is a much more affordable alternative for developing solutions with ChatGPT-like experiences. With a significant cost reduction in prices and a series of optimizations in the system, ChatGPT is poised to become a popular option for developers worldwide.