What is GPT3.5-Turbo-16K?

People know ChatGPT by its name, ChatGPT, but… those of us who have used the API to introduce ChatGPT into applications know that the name OpenAI gives it is “GPT3.5-Turbo”.

Now, recently “GPT3.5-Turbo-16K” has appeared in the playground in our OpenAI accounts.

This recent model is in principle the same as the “normal” 3.5 only that its context window has been extended to 16,000 tokens. This increases ChatGPT’s memory retention capacity by a factor of 8.

This update is very relevant 👍

To get an idea, approximately, ChatGPT is only able to remember a little more than 2 written pages, after that it does not remember the conversation and that is something that causes it to get lost. With this update it will be able to remember about 20 pages.

An impressive increase, but still far behind GPT4 at its best and Anthropic’s Claude who holds the record with 100k context window tokens.

As a curiosity, some users expected that having a larger context window, it would write longer answers, but this is not the case, the new version gives similar answers in terms of length, but it can remember more and therefore, through several requests, write a longer answer without losing coherence.