“Two versions of GPT-4 Turbo announced at once” – Kommersant FM

Kommersant FM columnist Alexander Levi talks about the capabilities of GPT-4 Turbo and prices for its users.

With more than 100 million weekly users of ChatGPT and 2 million developers using its API, the creators of generative chat, OpenAI, spoke about their achievements at their conference. There, the company presented a new flagship model – GPT-4 Turbo.

Naming is increasingly reminiscent of sports car names. In practice, the generative “gun-race” received a context window that is four times more powerful than its predecessor. The new model processes up to 128 thousand tokens or, in other words, generation directions in response to a request.

To understand, the Turbo version can fit about 100 thousand words or about half of the book “Harry Potter and the Prisoner of Azkaban” into the context window. This is the largest context window among all commercially available artificial intelligence models, Open AI representatives boasted.

A request of 1 thousand tokens or about 750 words will cost the user 1 cent. The generated result of the same volume already costs 3 cents. According to the calculations of the creators, the announced prices are approximately two to three times lower than before.

In general, the conference organizers announced two versions of GPT-4 Turbo at once. The first is intended for text analysis, the second recognizes the context of images. And in the latter case, the cost of artificial intelligence is determined based on the volume of information. For example, for 1 thousand images with a resolution of 1 million pixels it will be about $7.5.

The authors promised to make both versions of GPT-4 Turbo publicly available in the coming weeks.

The new model has access to information and events until April 2023, which updates and qualitatively improves generation results. GPT-4 Turbo also supports the so-called JSON mode, which simplifies interaction with files and data exchange.

Separately, it is worth noting the opportunity for users to create, publish and monetize their own GPT models through the OpenAI GPT store. The authors assure that this does not even require prior coding knowledge.