Llama

An open source large language model, from Meta. Llama 2 was pretrained on publicly available online data sources. The fine-tuned, pre-trained model, Llama Chat, leverages publicly available instruction datasets and over 1 million human annotations.

And Code Llama is a code generation model built on Llama 2, trained on 500B tokens of code. It supports common programming languages being used today, including Python, C++, Java, PHP, Typescript (Javascript), C#, and Bash.

As of April 2026, the Llama 4 family (released mid-2025) remains the primary open-weight flagship in Meta’s ecosystem. Llama 4 transitioned Meta from dense transformers to a Sparse Mixture-of-Experts (MoE) architecture, primarily to resolve the scaling bottleneck between parameter count and inference latency.