Meta Unveils Llama 3.2: New 1B to 90B AI Models with Multimodal Capabilities

Sep 25, 2024, 05:42 PM

Meta has unveiled its latest family of AI models, Llama 3.2, at Meta Connect 2024. The new models include lightweight versions with 1 billion and 3 billion parameters, designed for edge devices, as well as powerful multimodal models with 11 billion and 90 billion parameters. These models are capable of understanding both text and images and are positioned to rival leading closed models in the industry. The Llama 3.2 models are open-source, allowing developers to access and integrate them easily. They outperform OpenAI's GPT-4o-mini and Anthropic's Claude Haiku in various benchmarks. The models are available for download through platforms such as Hugging Face and GroqCloud, and they support several on-device use cases with partnerships from Arm, MediaTek, and Qualcomm. The Llama Stack simplifies development for developers and enterprises, and the models come with EU restrictions.