- OpenAI announced a new generation of embedding models and decreased prices on Thursday.
- The company is lowering prices on a new GPT 3.5 Turbo model, which powers ChatGPT.
- OpenAI is also unveiling a new GPT-4 Turbo model.
OpenAI is set to introduce a new GPT 3.5 Turbo model — the same model used to power ChatGPT — and will lower prices on the product for the third time in the past year.
The company announced the developments in a Thursday update, which included news of a forthcoming “new generation” of embedding models, as well as fixes to current models to address instances of “laziness.”
The new GPT-3.5 Turbo model will be introduced next week, OpenAI said, and both input and output prices for the model will be reduced: Input costs are decreasing by 50% to $0.0005 per thousand tokens in, and output costs are decreasing by 25% to $0.0015 per thousand tokens out.
The language model processes texts using tokens, which are frequent character sequences that exist in a text. GPT models teach themselves to understand the relationship between the tokens and can produce the subsequent token in a sequence, according to OpenAI.
The price cuts are intended to help OpenAI customers “scale,” the company said, and come as ChatGPT is increasingly being used to analyze larger vats of information and writing.
Lower prices could also be a result of the company’s efforts to streamline and improve models in recent months, especially as competition in the API community increases, TechCrunch reported Thursday.
The new GPT-3.5 Turbo model, which was named 0125 for the date of the announcement, is also set to include several product improvements, including higher accuracy at responding in requested formats and bug fixes, according to OpenAI.
GPT-3.5 Turbo has become the industry standard among other AI models seeking to compete with OpenAI’s accuracy and brand recognition.
OpenAI is also introducing an updated version of the GPT-4 Turbo preview model, the company said this week. While this model is more advanced than previous models in completing tasks like code generation, the new version will fix a prior bug in the GPT-4 Turbo that is meant to reduce cases of “laziness” in which the model wasn’t completing its task, the company said.
The update comes about three weeks after OpenAI reopened sign-ups for its subscription model, ChatGPT Plus, which is powered by GPT-4. Monthly subscribers to the service have access to more features, including the ability to interact with the chatbot by voice and engage in “back-and-forth” conversations with the product.