Llama 3.1 8B Instruct

Llama 3.1 8B Instruct is the instruction-tuned 8-billion-parameter release in Meta's Llama 3.1 family, launched on July 23, 2024. It extends the Llama 3 generation with a 128K-token context window, multilingual support, and native tool-calling and structured-output capabilities aimed at agentic use cases.

It became one of the most deployed open-weight models of its generation, serving as the default small model across inference providers and the base for thousands of community fine-tunes, including DeepSeek's R1 distill line. Weights are available under the Llama 3.1 Community License, which is a custom license with additional terms beyond standard open-source licenses, so commercial users should review it before deployment.