Llama 3.2 3B Instruct (NovitaAI)

Llama 3.2 3B Instruct is the instruction-tuned member of Meta's Llama 3.2 family, released under the llama3.2 community license and distributed through the meta-llama organization on Hugging Face. The underlying model card lists English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai among its declared languages, and the repo is tagged for the transformers text-generation pipeline, which is consistent with a small, general-purpose chat model designed for efficient on-device or budget-friendly server use rather than large-scale reasoning workloads.

In this catalog entry, the same Meta model is exposed as a hosted endpoint routed through the LLM Gateway with NovitaAI as the underlying provider, so practical strengths are best framed around that role. The gateway listing describes a text-only, temperature-controllable surface aimed at scalable text generation and natural language processing, with an open-weights license inherited from Meta even though this particular access path is closed. It is a sensible fit for lightweight assistants, summarization, classification, and other cost-sensitive multilingual tasks where a compact 3B parameter footprint is preferable to larger frontier models.