Meta has renewed its presence in the field of artificial intelligence with the Launch of Llama 4, an advanced model that is already active in applications such as WhatsApp, Instagram, and Messenger.Users of these platforms may notice improvements in the quality of responses from the virtual assistant identified by the familiar blue circle. This update is part of the company's strategy to consolidate its artificial intelligence ecosystem and compete with other major technology companies.
The deployment of Llama 4 has not been silentMark Zuckerberg himself announced its arrival, indicating that it is already available to many users and that all users need to do is update their applications to begin interacting with the new AI. Through his Instagram account, the CEO of Meta explained that this new generation promises a significant leap in speed, accuracy, and contextual understanding. This evolution in Meta's artificial intelligence is due to training with a huge amount of unlabeled information, including text, images, and video, which allows it to significantly improve its capabilities both for interacting with people and for performing complex tasks.
Where can you use Llama 4?
Llama 4 is now available on WhatsApp, Facebook Messenger, and Instagram Direct. and also through the official Meta AI website. If you want to check if you already have access, you can go to the AI section within the settings of each app. If Llama 3.2 is still listed, it's likely that your account hasn't been included in the initial rollout yet, as Meta is gradually activating it. However, if you experience issues, there are helpful tips for troubleshoot errors in WhatsApp.
It is important to note that this feature cannot be completely disabled from the platforms., although its impact can be minimized by deleting the AI chat and typing the command “/reset-ai”, which deletes the conversation history associated with the assistant.
This new version improves Meta AI's overall performance in multiple aspects, from text generation to image interpretation. Key areas of benefit include:
- Faster and more contextualized responses
- Better understanding in multiple languages
- Ability to write texts, create code or develop ideas
- More fluid and natural interaction
In addition, Meta has opted for offer different models of Llama 4, tailored to different contexts and needs. This variety allows performance to be tailored to available resources and usage requirements.
The four variants of Llama 4
In total, Meta has presented four versions of the Llama 4 model, two of which are already active and available, while the other two are still in the development phase:
Llama 4 ScoutThis model stands out for its speed and multimodal capability from its initial design. It has 17.000 billion active parameters and 16 experts. Its context window reaches up to 10 million tokens, allowing it to analyze text and images from large documents. It is optimized to run on a single GPU, making it accessible for resource-limited environments.
Llama 4 MaverickDesigned to function as a general assistant or versatile chatbot, Maverick also features 17.000 billion parameters, distributed across 128 experts. According to Meta, this version outperforms popular models like GPT-4 or Gemini 2.0 in certain cases when testing logic, coding, multilingual comprehension, and handling extended context.
Llama 4 ReasoningThis variant, still pending release, will promise advanced logical reasoning capabilities. It is specifically designed for tasks that require deeper and more reliable processing of complex information.
Llama 4 BehemothMeta considers this model to be the most powerful ever developed within its lineup. It's an AI with over 2 trillion total parameters, of which 288.000 billion are active. Although it's still in the training phase, it's expected to redefine what can be achieved in open AI.
Where is it available?
Currently, Llama 4 is operational in 40 countries, although there are notable exceptions.For example, its use is not yet permitted within the European Union due to certain legal restrictions related to data protection and artificial intelligence regulations.
Furthermore, Llama 4's advanced multimodal features are currently only available in the United States and only in English.This means that many users outside the US still cannot access combined text and image analysis, one of the key capabilities of the new model.
One of the most relevant new features of Llama 4 compared to Llama 3 is its architecture based on the “mixture of experts” system.This structure allows tasks to be divided into subprocesses, which are then assigned to specialized models. This results in greater efficiency in both training and execution, without sacrificing results.
It is also noteworthy that Meta has adjusted the assistant's response filters.While in previous versions, some sensitive questions were avoided, the model is now expected to provide more neutral and balanced answers, even when dealing with social or political issues.
As the company explained, these adjustments seek to improve the real usefulness of AI in everyday conversations., while avoiding ideological extremes or marked biases.
The Scout and Maverick models are available for download on platforms such as Hugging Face and the official website llama.comThis allows developers and researchers to experiment with them directly or integrate them into their own technology projects.
The arrival of Llama 4 marks a new step in the development of large-scale open artificial intelligence. It features a focus on accessibility and adaptability. The different versions allow each user or company to choose the model that best suits their needs. For example, simple tasks in mobile apps or even complex implementations on specialized servers.
