The Latest ChatGPT Update: Uncovering Its Unexpected Changes and Features

OpenAI has announced the rollout of an updated version of its GPT-4o large language model, which powers the ChatGPT chatbot. However, the company has not detailed how this new version differs from its predecessor. “To clarify, this is an enhancement to GPT-4o, not a new frontier model,” the company stated on X (formerly Twitter) on Monday.

According to the Model Release Notes, OpenAI noted, “We’ve introduced an update to GPT-4o that has shown, through experimental results and user feedback, to be preferred by ChatGPT users.” The company emphasized that while they are eager to share the specifics of how model responses differ, they are still researching ways to benchmark and communicate improvements to model behaviors effectively.

In the absence of clear information, many users have started speculating about the nature of the changes and whether they introduce any new features. One user on X, @misaligned_agi, suggested that the update might have implemented a multistep reasoning method rather than introducing an entirely new model.

OpenAI quickly addressed this speculation, with a spokesperson clarifying to VentureBeat that the update doesn’t involve a new reasoning process and that @misaligned_agi's observations may have been influenced by the wording of their prompts.

Other users shared their thoughts online, noting that GPT-4o appears to have improved in subtle ways, especially in image generation quality. User @mattshumer_ commented, “For the first time in a long time, it provided better ‘vibes’ on an output than 3.5 Sonnet.”

After users made their guesses about the updates, OpenAI provided additional insights on its Models page on Wednesday. The updated model, referred to as ChatGPT-4o-latest, is described as a "dynamic model continuously updated to the current version of GPT-4o in ChatGPT." It maintains a knowledge cutoff of October 2023 and can handle 128,000 tokens, or 96,000 words, per conversation, similar to the earlier GPT-40 version. It can generate 16,384 tokens, or 12,288 words, which aligns with the capabilities of the newer GPT-40-mini model, and is roughly four times what the previous GPT-4o could deliver. Unfortunately, these specifications don't reveal much about the new model's actual capabilities, and it seems that OpenAI will not disclose further details at this time.