OpenAI introduced a more powerful version of GPT-4 Turbo and reduced the price of APIs for language models

D koi/Unsplash

OpenAI has announced enhancements to its GPT-4 and GPT-3.5 language models. The new versions include updated knowledge bases and a longer context window.

Here's What We Know

GPT-4 Turbo, available as a preview version of the API, has been trained on data until April 2023. Early GPT-4 only used data until September 2021. A full release of Turbo is planned for the coming weeks.

GPT-4 Turbo also sees more context - 128 Kbytes of text at a time. This will allow the model to better understand questions and provide thoughtful answers. Previous versions had 8K and 32K context.

Turbo is also cheaper for developers - $0.01 per 1,000 tokens on entry and $0.03 on exit. That's three times lower than the previous price, OpenAI claims.

GPT-4 will still accept text queries, image hints and integrate the DALL-E 3 image generation function. A single query can be asked to produce the result in the desired format: XML, JSON and so on.

The smaller GPT-3.5 Turbo model with a 16 Kbyte context window will get similar improvements. Its cost will be $0.01 on entry and $0.002 on exit.

In addition, OpenAI has promised to reimburse customers for copyright infringement lawsuits, similar to Microsoft and Google. This applies to ChatGPT Enterprise and the developer platform.

Source: OpenAI