THE DECODER

Meta has released Llama 3.3, a new AI model that matches the capabilities of its larger Llama 3.1 405B while using significantly less computing power. The company says the model excels at tasks such as generating synthetic data while keeping inference costs low. The improvements come from a new alignment process and advances in online reinforcement learning techniques, according to Meta. The model is now available through Hugging Face and will roll out to partner platforms soon. Meta notes that Llama 3.3's efficient design allows it to run on standard developer workstations without requiring specialized hardware.