OpenAI introduced an improved GPT-4 Turbo model. She can analyze an entire book

OpenAI announced a new model GPT-4 Turbo, which has become more powerful and cheaper. Its knowledge base was updated with world events up to April 2023, while GPT-4 only knows about events until September 2021.

We also increased the input data limit to 128 thousand input tokens - this is approximately 300 pages of printed text, or a medium-sized book. The additional context allows language models to better understand the request and provide a more detailed response. Basic GPT-4 can handle requests for up to 8 or 32 thousand tokens depending on the version.

In addition to text, the model also receives requests with pictures, voices responses and writes code. With GPT-4 Turbo, the user can even choose the programming language when asked to write or analyze code.

GPT-4 Turbo is currently only available as a preview API for developers. It has become cheaper to use: a thousand input tokens now cost only 1 cent, and output tokens - 3 cents. GPT-4 costs 3 and 6 cents, respectively.

Conditions have also been improved for GPT-3.5 Turbo users. By default, the model will work with requests up to 16 thousand tokens in length and will receive new GPT-4 Turbo features. A thousand tokens will cost 0.1 cents for entry and 0.2 cents for exit.