OpenAI launched GPT 4 Turbo in November last year, shortly followed by the GPT Store as well as custom GPTs. Now the big update, called GPT 4.5 Turbo, could launch as soon as June 2024, as per a new leak.
The leak comes from none other than OpenAI’s own web team. Before any formal unveiling, initial search engines such as Bing and DuckDuckGo had already cataloged the product page for GPT-4.5 Turbo.
Despite the indexed link directing users to a 404 error page, a snippet of teaser text in the search results reveals that GPT-4.5 Turbo is touted as OpenAI’s quickest, most precise, and most expansively scalable model yet. The specific details behind these claims are still a mystery.
OpenAI is expected to unveil an official launch date tomorrow on Thursday, which just so happens to be the anniversary date for GPT 4. OpenAI’s CEO, Sam Altman, is scheduled to appear on Lex Fridman’s podcast, mirroring a previous session that took place following GPT-4’s release.
GPT 4 Turbo already introduced a major step forward in terms of text input lengths, boasting a context window of 128,000 tokens. But GPT 4.5 Turbo is expected to take this much further with double the context length sized at a whopping 256,000 tokens, which is about 200,000 words. This would make it ideal for comprehensive documents, research papers, etc.
However, the key performance indicator, more important than the volume of input data, is the model’s ability to accurately process such data. Present models featuring extensive context windows often overlook information, diminishing their utility in analyzing large text documents.
Should GPT-4.5 Turbo exhibit improved performance in this area, it would represent a significant advancement, even if its overall performance mirrors that of existing models. Google, too, is in the process of beta-testing a potent large context model, Gemini 1.5.
GPT 4.5 Turbo will also update the AI model’s knowledge cut-off date to June 2024, according to the teaser’s description. This could mean that GPT 4.5 Turbo will launch around June, which would be unusual for OpenAI since the company is known for making its products available soon after launch.
