GPT-4.5 release rumors are heating up. Will OpenAI's answer to Google Gemini be released in December?
The leak account "Jimmy Apples" last speculated in early December on X about a possible release of GPT-4.5 at the end of December. Now, the first users report that ChatGPT already answers the question of which model it uses with "gpt-4.5-turbo" - this is also the case for me.
This could also be a hallucination, because if you ask about the differences between "gpt-4.5-turbo" and "gpt-4.0-turbo", you will only get generalities based on knowledge from April 2023.
It would also be unusual for OpenAI to release the turbo version of GPT-4.5 directly. For GPT-3.5 and GPT-4, the turbo versions were released months later.
In mid-December, a screenshot of alleged pricing for GPT-4.5 surfaced, showing that the new model would support audio, video, and 3D in addition to text and images. It is supposed to solve complex logic problems and feature cross-modal understanding.
Notice in the screenshot that there is no arrow at the end of the link to GPT-4.5 - this could be a mistake made by the faker, or someone forgot to check a box in the content management system.
GPT-4-Turbo was also leaked in advance via a price table on the website. However, OpenAI CEO Sam Altman called the 4.5 table a fake.
OpenAI could enhance GPT-4's multimodality with video and 3D
The fact that the screenshot leak may be fake does not mean that GPT-4.5 Turbo is not in the wings. The rumors fit with previous reports of AI capabilities that OpenAI is currently working on.
Back in March, Microsoft Germany CTO Andreas Braun said that OpenAI was working on multimodal models with video generation. These rumors were repeated by OpenAI leaker FeltSteam in connection with "Project Arrakis," which is said to offer video generation, have fewer hallucinations than GPT-4, and be suitable as an autonomous agent.
Although Project Arrakis, which was designed to be less expensive than GPT-4, reportedly failed, the work may have been incorporated into new models. One candidate is Project Gobi, another AI prototype that will support video and could become GPT-4.5 or GPT-5, according to early reports. The new model could be released in early 2024.
With Point-E and Shap-E, OpenAI has already introduced two AI systems that can turn text into 3D models.
In competition with Google, it would be characteristic of OpenAI to go one better shortly after the Gemini presentation. Microsoft has already countered the Gemini Ultra presentation with optimized GPT-4 benchmarks.