OpenAI of GPT-4.5Turbo appears to have been leaked, with search engines like Bing and DuckDuck Go indexing the product page for GPT-4.5Turbo before the official announcement.
However, the indexed link points to a 404 page. But there is some teaser text in the search engine, saying that GPT-4.5Turbo is OpenAI's fastest, most accurate, and most scalable model to date. What exactly these terms mean remains to be seen.
It is predicted that the new model will likely be officially unveiled this Thursday, coinciding with the one-year anniversary of the release of GPT-4. OpenAI CEO Sam Altman is also expected to appear on the Lex Fridman podcast, following their in-depth conversation following the release of GPT-4. GPT-4.5Turbo: Greater processing range andup to dateThe “context window” mentioned in the information preview text refers to the amount of data that the model can process at one time. GPT-4.5Turbo can process up to 256,000 tokens, which is twice the processing power of the previous GPT-4Turbo128K, which is equivalent to approximately 200,000 words.
However, more important than the amount of input data is how reliably the model can handle all of it. Current models with large context windows tend to ignore information, making them less valuable for text analysis of large documents. If GPT-4.5Turbo performs better in this regard, it will be a big win for the model even if performance is similar to current models in other respects. Google is beta testing Gemini1.5, which has a powerful large context model.
The preview description also indicates that GPT-4.5Turbo will be updated until June 2024, the so-called "knowledge deadline". This suggests that the GPT-4.5Turbo model may be released in June. This may be quite unusual for OpenAI: so far, newly launched models have been available immediately after they are launched.
Rumors about the upcoming release of GPT-4.5Turbo date back to at least December 2023. Rumors suggest the new model may have video or 3D capabilities in addition to text and images. However, the leaked preview did not mention multimodal capabilities. A few days ago, OpenAI CEO Altman hinted at a new AI model that is "worth waiting for."
In addition, a recent announcement from the University of California, Irvine (UCI) shows that they are testing GPT4.5. Their own chatbot ZotGPT, based on Azure AI, provides free access to GPT-4.5, supports chat history, file uploads, and large context windows. This may also hint that GPT-4.5 Turbo may be coming soon.