OpenAI has announced a new iteration of its language model GPT-4 in a short event streamed this afternoon. GPT-4o, that's what this version is called, is “Too much faster” than that and is natively multimodal. That is, it can understand and generate text, images and audio, as explained by the CTO of OpenAI, Mira Muratyisomething of which several examples have been seen during the presentation.
Surprisingly, the new language model It will not be reserved for users of the paid version, but reaches all ChatGPT users, also in the free version. What ChatGPT Plus users will be able to enjoy is “up to five times the capacity limits” available to ChatGPT Plus users and they will maintain exclusive control over the voice capabilities of the new model. While the current voice mode works only with what it can hear, GPT-4o can observe what's around it through the camera and talk about it; and your responses will be more agile too. In this case, the latency has been reduced from 5.4 seconds for GPT-4 and 2.8 for GPT 3.5 to 320 milliseconds, similar to the response time in a conversation between humans.
OpenAI has shown some of the capabilities of the new model with spoken conversations that inevitably reminds us of the AI assistant in the movie Her starring Joaquin Phoenix. The one from OpenAI seems to dialogue naturally, interpret the user's emotions and adapt its style to what they ask of it, in addition to incorporate sound effects such as laughter into the conversation.
The company has noted, in a blog post, that GPT-4o's capabilities “will be implemented iteratively (with extended access to the red team starting today)”, but its text and image capabilities start coming to ChatGPT today.
The new version of ChatGPT reduce latency, which will allow faster responses and is optimized for more than 50 languages. As indicated by Sam Altman, CEO of OpenAI, in X, developers will have access to its API for their own implementations and will be able to do it at half the price than with GPT-4 Turbo, the now penultimate version of the language model that is available to ChatGPT Plus users; GPT 3.5 is used in the free ChatGPT. Altman has also noted that GPT-4o “It is our best model to date“.
in the API, GPT-4o is half the price AND twice as fast as GPT-4-turbo. and 5x rate limits. pic.twitter.com/vqV8XwNcYp
—Sam Altman (@sama) May 13, 2024
The OpenAI announcement had sparked a lot of speculation in recent days, so much so that Altman had to come forward and ensure that It would not be GPT-5 or a search engine with AI.
A new iteration of GPT-4 is not what many expected, but the move to bring it to all its users is a novelty in terms of ChatGPT's business model. Muratyi has also advanced the launch of a ChatGPT desktop app, currently for Macand an update to the user interface of the web version of ChatGPT.