Meta is launching the latest version of its LLM Llama 3.2 with multimodal features. However, these are not available in Europe: we have to make do with limited versions.

During the Connect developer conference, where new VR and AR headsets were also unveiled, Meta announced the latest versions of the LLama model. The latest version, Llama 3.1, was barely two months old when Llama 3.2 was already a fact. Google and OpenAI regularly announce new variants of their models, so Meta needs to make sure they keep up.

Meta’s Llama 3.2 is released in several variants. The most extensive model has 90 billion parameters. Meta claims that LLama 3.2 can process up to 100,000 words in a single prompt. The most important new feature is the multimodal capabilities, which now allow LLama to process images and text simultaneously, something that was not possible with previous versions.

This allows your llama to show a photo or graphic and ask questions about it in the text. The model can interpret diagrams and perform analysis or mark certain objects in a photo. The 11B version with eleven billion parameters also has multimodal capabilities.

Not for Europe

This all sounds very nice, but unfortunately it will not be of any use to us in Europe. Meta is not bringing the 90B and 11B variants to our region. This is no surprise, as Meta had already more or less announced this. To compensate, we will get the more compact versions with one and three billion parameters, but these are limited to text.

Meta is imposing this restriction on Europe to make a statement. Mark Zuckerberg’s company wanted to collect data from European users through its social media platforms, but had to pull its tail after protests from data protection organizations. Meta not only has problems with the GDPR law, but is also not a big supporter of the AI ​​law.

Llama 3.2 is available through Hugging Face, Google Cloud, Microsoft Azure, AWS, or directly from the Meta website. The model is open source, meaning the code is not kept behind closed doors. The openness of Llama models is a key differentiator for Mark Zuckerberg, and Zuckerberg is a firm believer that open source will win out over “closed” models.