GPT-4 Turbo

What it is

GPT-4 Turbo is a version of OpenAI's GPT-4 language model that the company described as cheaper and more capable than GPT-4 ^[1]. It was announced at OpenAI's first DevDay on 6 November 2023 ^[1][3]. Rather than a wholly new kind of model, it was a revision of the existing GPT-4 line aimed at the developers who build on OpenAI's API: the same family of capabilities, but with a larger working memory and a lower price per token.

A much larger context window

The most visible change was the size of the context window. GPT-4 Turbo accepts 128,000 tokens in a single request, which OpenAI said fits about 300 pages of text in one prompt ^[1][6]. A context window is the amount of text a model can hold at once — both the instructions it is given and the material it is asked to work with. A prompt that can carry roughly 300 pages means a whole book, a long source file, or a stack of documents can be handed to the model in one go, instead of being cut into pieces and processed separately.

The second constraint is the knowledge cutoff. OpenAI said the model had knowledge of world events up to April 2023 ^[1][6]. A cutoff is simply the point where the training data ends: the model has no built-in awareness of anything after that date, so recent information has to be supplied in the prompt or fetched by other tools.

Price

For the 128K model, OpenAI priced GPT-4 Turbo at $0.01 per 1,000 input tokens and $0.03 per 1,000 output tokens ^[2]. Input tokens are the text sent to the model — the prompt, the documents, the conversation so far — while output tokens are the text the model generates in reply. Charging them at different rates reflects the fact that generating text is the more expensive operation, and it gives developers a direct incentive to keep prompts lean and answers focused. The lower headline price is what made the word "cheaper" in OpenAI's description concrete: the same budget buys more requests than it did with GPT-4 ^[1].

The DevDay setting

GPT-4 Turbo was not announced alone. The same event introduced the Assistants API, multimodal capabilities for the platform including vision, DALL·E 3 image creation and text-to-speech, and lower prices across much of the platform ^[6]. Read together, those announcements describe a shift in emphasis: the model itself was one item on a list that also covered the tools wrapped around it, the kinds of media it could handle, and the cost of using it.

Versions and shutdown

Developers reached GPT-4 Turbo through several API model names: gpt-4-1106-preview, gpt-4-0125-preview and gpt-4-turbo ^[4]. Naming model versions this way lets an application pin itself to a specific snapshot of the model rather than to whatever is current, which matters when a product depends on stable behaviour. The other side of that arrangement is that older names get retired. gpt-4-0125-preview, together with its gpt-4-turbo-preview alias, was shut down on 26 March 2026, while gpt-4-turbo and gpt-4-1106-preview were scheduled for shutdown on 23 October 2026 ^[4]. Applications built on those names had to move to a newer model before those dates.

Place in the line

GPT-4 Turbo came after GPT-4 and was itself followed by GPT-4o in May 2024, which OpenAI said matched GPT-4 Turbo on English text and code while being 50% cheaper in the API ^[5]. That sequence is a fair picture of how the field moved in this period: each generation was measured against the previous one on capability, on how much text it could handle at once, and on price per token. GPT-4 Turbo is the step where a very large context window and a much lower price arrived together, and where the surrounding platform — assistants, images, speech — was bundled into the same release.