Meta will release in the coming weeks the new version of its generative AI model, call 4, and according to Financial Times will be centered on improve your voice capabilities.
The star development of Zuckerberg’s firm in artificial intelligence has accelerated after Depseek’s emergence. American media have that Meta executives rushed to establish operations rooms to decipher how Chinese development had reduced the execution and implementation costs of their models and at the same time achieved yields similar to the large models of the sector. Something similar has happened with OpenAI and the accelerated deployment of GPT-4.5.
Call 4 will continue to bet on a type of “open” developmentwhich means that developers can download it and use it as they want (with certain limitations). Although it does not reach the Open Deepseek source model, in this section it goes beyond closed models such as the Claude of Anthropic, the OpenAI GPT-4o (which drives Chatgpt) or the Google Gemini, which can only be accessed through an API interface.
In addition, with the aim of offering greater Options to developersMeta has also been associated with suppliers such as AWS, Google Cloud and Microsoft Azure to make flame versions hosted in the cloud and it has support for a variety of hardware platforms offered by AMD, Dell, Intel, Nvidia and Qualcomm. The company has also launched additional tools designed to facilitate the adjustment and customization of the model.
Entering the subject, call 4 will be the next flagship of the LLM family of Meta and promises advances in all sections. Like other generative AI models, you can perform a variety of different assistance tasks, such as coding and answering basic mathematical questions, as well as summarizing various languages. Although it can be configured to take advantage of applications, tools and third -party API to complete tasks, so far it has been focused on text mode.
This will change with flame 4 since the main improvements will be destined to Train it for voice use. According to the director of Meta products, Chris Cox, it will be an “omnidirectional” model, capable of interpreting and generating in a native voice, as well as text and other types of data. Goal has focused especially on allowing users to interrupt the model in the middle of their speech, similar to Openai’s voice mode for Chatgpt and Google’s Gemini Live experience.
And it is that, like other AI giants, he believes that The future of agents promoted by AI will be conversational Instead of text based, although to date a minimum degree of the naturalness and bidirectionality of humans has not been achieved. Goal wants users to interrupt the AI model and have a more fluid dialogue, instead of a rigid format of questions and answers.
Call 4 will be published in “The coming weeks” And it will be the basis to offer agent artificial intelligence “Hundreds of” of companies, as Clara Shih, the Director of Artificial Business Intelligence of Meta, says,
