CodeLlama 13B

CodeLlama 13B is Meta's 13-billion-parameter general code generation model, released August 24, 2023. It is produced by further training Llama 2 13B on code-heavy data and supports code completion, fill-in-the-middle (FIM) infilling, and code understanding across multiple programming languages. The listing carries a 100K-token context, so it should be read as a long-context hosted listing for the 13B base model rather than a guarantee that every CodeLlama 13B endpoint exposes the same window.

At 13 billion parameters, CodeLlama 13B offers a meaningful capability improvement over the 7B variant: stronger handling of complex multi-function code, less common programming languages, and more accurate code completions for longer functions. The model covers Python, C, C++, Java, PHP, TypeScript, C#, and Bash, among others. The 100K context row is most relevant when a provider exposes extended code context for larger files, repository excerpts, or long autocomplete histories; verify the selected provider's cap before designing around full-window prompts.

CodeLlama 13B is available as open weights on Hugging Face (meta-llama/CodeLlama-13b-hf) under Meta's Code Llama Community License, and is hosted on Together AI, Fireworks AI, Azure AI Foundry, and Replicate. For Python specialization, the CodeLlama 13B Python variant is preferred. The 34B and 70B variants offer higher accuracy at greater compute cost.