What is GPT-4 Turbo? The successor to GPT-4 that's available in preview now

OpenAI recently unveiled GPT-4 Turbo, and the only way to actually use it as a normal consumer currently is through Microsoft Copilot. Developers can use it through OpenAI's GPT-4 API by passing "gpt-4-preview" to the API in the request, but normal users will either have to wait or switch to Copilot.

What does GPT-4 Turbo have though that regular GPT-4 doesn't? As the name would imply, it's an incremental improvement, but still an improvement nonetheless.

GPT-4 versus GPT-4 Turbo

Minor changes, but ones that users will appreciate

The biggest leap that many will experience in GPT-4 Turbo compared to GPT-4 is its extended knowledge window, updated to include events up to April 2023. This improvement significantly benefits users interested in researching more current events, as the model can now offer up-to-date information on occurrences up to that point. That's not all though, as there is also a significantly larger context window to play with.

For the uninitiated: context windows in AI are windows that compromise the entirety of the AI's "thinking." This includes all the input it uses to understand and generate a response, encompassing the prompt given by the user as well as any additional context or previous dialogue. The size of the context window determines how much information the model can "remember" from earlier in the conversation or document, affecting its ability to produce coherent and relevant responses. GPT-4 Turbo increases this to 128K tokens, with a maximum response token length of 4,096. This is a 4x increase in context window from GPT-4's 32K window, which is a huge improvement.

For developers, GPT-4 Turbo is also significantly cheaper, as it's cheaper for OpenAI to operate. When compared to GPT-4, it's 3x cheaper in price for input tokens and 2x cheaper in price for output tokens. OpenAI also says that it's better at following specific instructions, such as telling it to "always respond in XML format", and supports a new "JSON mode" so that it can respond with valid JSON for developers that are incorporating it in their applications.

OpenAI has been publishing changelogs as it improves GPT-4 Turbo, and the most recent changelog states that it is "ntended to reduce cases of “laziness” where the model doesn’t complete a task." This is in response to a recent phenomenon of people stating that GPT-4 felt "lazy" because it would instead tell users how to complete a task, rather than doing it when asked. This led to some pretty funny workarounds, including telling ChatGPT that you would give it a sizable tip for a good answer.

All in all, though, GPT-4 Turbo can do all of the same things as regular GPT-4 can, it's just cheaper for OpenAI to run and has a larger context window. It apparently has better responses as well, but "better" is often suggestive, and as it's in a preview stage right now, it's hard to say whether it actually is better or not. It'll evolve over time, though, and we'll be able to say more in the future.

How to use GPT-4 Turbo

Only available for developers right now... or through Copilot Pro

If you want to use GPT-4 Turbo via ChatGPT, you'll need to do it through OpenAI's API. There's no other way to access it right now, though you can also purchase a Microsoft Copilot Pro subscription and use it through that service instead. It's not quite the same, though many would argue that Copilot Pro is actually better than ChatGPT Plus is anyway.

Alternatively, you can use Microsoft Copilot for free and get to access GPT-4 Turbo at off-peak hours. Copilot Pro will give you priority access during peak times, but you can still try it out otherwise. It's being rolled out to users, so if you don't have it yet, you likely will soon. Microsoft said it would arrive "in the coming weeks" in December, so it's very likely that you'll have it soon if you don't already.