OpenAI announces major API updates such as 'improvement of GPT-4' and 'reduction of costs', and is also enthusiastic about increasing the number of people who can use GPT-4
OpenAI, which provides chat AI such as ChatGPT, has announced an update that significantly improves API pricing and functionality. The content further expands the possibilities of products incorporating chat AI.
Function calling and other API updates
The improved contents are as follows.
・Function call function added to Chat Completions API
Chat Completions API is an API that generates responses based on chat history, and many people should associate this with chat AI.
With this update, gpt-3.5-turbo and gpt-4 models can now include function information in the input. For example, if you prepare a function that obtains weather information and send that function information along with the user's conversation data to the Chat Completions API, when the user asks, 'What is the weather like in ○○?' The data 'Call this function with this argument!' is returned. By sending the execution result of that function from the application side to the Chat Completions API again, an appropriate response is generated.
The existing model will not be updated until June 27th, two weeks later, but if you want to use the function call function now, add '0613' to the model name. For example, in the case of 'gpt-4', specifying the model with the name 'gpt-4-0613' allows access to the model with the function call function.
・New model 'gpt-3.5-turbo-16k' appeared
The conventional 'gpt-3.5-turbo' model could handle up to 4096 tokens, but the new 'gpt-3.5-turbo-16k' model can handle up to 16,384 tokens. The price is $ 0.003 (about 0.42 yen) per 1000 input tokens and $ 0.004 (about 0.56 yen) per 1000 output tokens.
・Model replacement timeline
If you access only the model name such as 'gpt-3.5-turbo' or 'gpt-4', the contents will be changed to the new version on June 27th. If you want to continue using the existing version, it is OK if you specify 'gpt-3.5-turbo-0301', 'gpt-4-0314', 'gpt-4-32k-0314' as the model name. However, it has been announced that the use of these older models will end on September 13th.
・Price drop
The price of the model 'text-embedding-ada-002', which converts text into numerical values and can be used for searching and similarity comparison, is reduced by 75% and available at $ 0.0001 (about 0.014 yen) per 1000 tokens became.
The fee for the 'gpt-3.5-turbo' model has dropped by 25% to $0.0015 (about 0.21 yen) per 1000 input tokens and $0.002 (about 0.28 yen) per 1000 output tokens. It's just half the price of the 'gpt-3.5-turbo-16k' model.
In addition to these model updates, the blog post stated that the GPT-4 waiting list will be reduced over the next few weeks, and will eventually be removed entirely.
Related Posts:
in Software, Web Service, Posted by log1d_ts