Browse Models
The Llama 4 family, developed by Meta and released in April 2024, comprises natively multimodal large language models featuring mixture-of-experts transformer architectures with 17 billion active parameters per token. The models support unified text and image processing, long-context understanding up to 10 million tokens, and multilingual capabilities across over 200 languages, with training data including approximately 40 trillion multimodal tokens and knowledge cutoff in August 2024.