GPT-3.5 and GPT-4 can now develop code based on a description of programming features. Even better news: the price per token will be reduced.

OpenAI announces innovations for GPT-4 and the “turbo version” of the predecessor GPT-3.5. In addition to increased “controllability”, the coding capacities of the models are also increasing. With a new feature called function call The developer can describe a programming function and GPT does the rest.

Based on this description, the model then knows which plugin needs to be called. The AI ​​extracts the necessary data from your text and converts natural language into JSON, a commonly used “language” to get computer programs to “talk” to each other to perform the function. For more technical explanations or examples of how this process works, see this OpenAI blog.

More tokens, pay less

In addition, OpenAI also extends the maximum number of “tokens” that GPT-3.5 Turbo can handle from 4,000 to 16,000. GPT-4 has a limit of 32,000 tokens to clarify the difference. The GPT models convert characters from our language into tokens. The number of tokens that GPT counts in a word or phrase does not automatically match the number of characters you type. We would like to refer you again to a blog where you can read more about this mechanism.

OpenAI also has the good news that developers can now get more tokens for less money. The original turbo version of GPT-3.5 now costs you $0.0015 per thousand input tokens and $0.002 per thousand output tokens. The version that can handle 16,000 tokens costs $0.003 and $0.004 per token. It text-embedding-ada-002 The price drops 75 percent to $0.0001 per thousand tokens. Unfortunately, we cannot calculate for you how much it would cost you to use the paid APIs.