Call 4 is the New generation of large languages (LLM) of finish. Mark Zuckerberg’s company has just announced its launch this weekend and will arrive in three different versions.
The battle for AI continues at high and language models, specifically designed to work with the natural language that humans use to communicate, are key because of the large number of applications and services that use them, integrating their artificial intelligence functions.
Goal calls 4
Goal says that smarter systems must be able to perform generalized actions, communicate naturally with humans and “Solve challenging problems that have not seen before”. Hence the evolution of the LLM promises as a result “Better products for users of our platforms and more opportunities for developers to innovate in the next large cases of use for consumers and companies”.
Goal is distributing three different versions which are mainly differentiated by the number of parameters (equivalent to the neuronal connections of a neuronal network) that they can use.
Call 4 scout It is the smallest model in the series, with 17 billion active parameters and 16 experts. Meta states that Scout is the best multimodal model in its class, surpassing Gemma 3, Gemini 2.0 Flash-lite and Mistral 3.1 in standard performance tests. Despite its performance, this model can be executed in a single professional GPU of the NVIDIA H100 series. It should be noted that it admits a context window of 10 million tokens. Really impressive, although the real effectiveness of such a wide window is yet to be demonstrated.
Call 4 maverick It is the main model, with 17 billion active parameters, but scalable to 128 experts. Meta states that Maverick is the best multimodal model in its category, surpassing the widely used GPT-4O and Gemini 2.0 flash in the sector’s reference tests. His experimental version of Chat obtained a score of 1417 in Lmarena, occupying second place among the main LLM.
Llama 4 Behemoth It will be the largest model of the line, but it is still in the development phase. It has 288 billion active parameters and 16 experts. According to Meta, this massive model exceeds GPT-4.5, Claude Sonnet 3.7 and Gemini 2.0 Pro in several standard AI performance tests.
Call 4 scout and call 4 maverick are now available to download. For the general public, these models incorporate goal AI in company applications such as WhatsApp, Messenger, Instagram Direct. They are also available via web and third companies such as Microsoft have announced their incorporation in Azure AI Foundry as administered computer solutions.
