The rumors were true. A few days ago, OpenAI officially unveiled “Project Strawberry,” which turned out to be a new version of their language model GPT-4o. “There’s a new GPT-4o model in ChatGPT since last week,” the company announced in a tweet published on Monday. “We hope you all enjoy it and, if you haven’t already, take a look: we think you’ll like it.” However, OpenAI has not provided any details about the new features that this model brings.
In a response to another tweet, OpenAI clarified that the new GPT-4o model is available for both paid subscription users (ChatGPT Plus) and those using the free version, although with limitations on the number of messages. Additionally, the company confirmed that the new GPT-4o is not the same as the GPT-4o-2024-08-06, released last week.
Before the official announcement, several ChatGPT users already noticed differences in the chatbot’s behavior. According to a report from VentureBeat, some users perceived that the model was functioning more efficiently and that the capabilities of native image generation seemed to have been activated. Others highlighted an improvement in multi-step reasoning, a technique that breaks down complex problems into smaller, more manageable steps.
However, OpenAI denied that it was an improvement in multipass reasoning. Instead, the company described the update as a series of “bug fixes and performance improvements” that ChatGPT users have preferred, according to the results of several experiments and qualitative feedback. Nevertheless, OpenAI admitted that it is still difficult to accurately communicate how the new model behaves.
Despite the lack of clarity about the exact improvements, the new model, called ChatGPT-4o-latest, has proven its worth. In tests conducted by Chatbot Arena, this model ranked as the best, outperforming its competitors, such as Google’s Gemini 1.5, with outstanding scores in technical areas such as coding, instruction tracking, and solving difficult questions.
