Is there a new version of ChatGPT in the waiting room? A leaked blog from OpenAI announces the premature birth of GPT-4.5 Turbo.

OpenAI may have accidentally announced the launch of a new version of the GPT model. It would be called GPT-4.5 Turbo and is the successor to GPT-4 (Turbo). The announcement shouldn’t go live now as OpenAI’s blog was quickly removed. But by then the damage was already done: the blog was indexed by the search engines DuckDuckGo and Bing, and screenshots were circulating widely on Reddit and X.

Higher token limit

Some details about GPT-4.5 Turbo can be found in the short snippet. The new version would have knowledge of current events up to and including June 2024, making a possible summer launch a realistic option. The current state of knowledge about GPT-4 ends in April 2023.

The biggest improvement over GPT-4.5 Turbo would be the number of tokens the model can process per prompt. It’s not illogical that OpenAI is working on this: Google’s Gemini and Anthropic’s Claude have overtaken GPT in this area. GPT-4.5 Turbo would increase the token limit to 256,000 tokens, twice what GPT-4 can handle. This corresponds to 150,000 to 200,000 words. GPT-4.5 Turbo can therefore process very long texts (or media files) and filter out very specific information.

Note from the co-pilot?

OpenAI doesn’t want to confirm anything about GPT-4.5 Turbo for now, but with the leak now open and exposed, it will have no choice but to communicate quickly. Another indication that a new version of GPT is on the way comes from the Microsoft camp. The sister company is adding GPT-4 Turbo to the free version of its Copilot app, meaning there are virtually no fees to use the smartest version of GPT. There have also been rumors for some time that OpenAI would be working on GPT-5, but for now the company seems to see enough scope for GPT-4.