Llama 3.2 3B Instruct offers capability gains over the 1B, outperforming comparable small models on summarization, instruction following, and tool use.
import { streamText } from 'ai'
const result = streamText({ model: 'meta/llama-3.2-3b', prompt: 'Why is the sky blue?'})Meta released Llama 3.2 3B Instruct on September 18, 2024 as the larger of its two lightweight text-only models in the Llama 3.2 collection. At 3B parameters, it consistently outperforms its smaller sibling across evaluation tasks.
Llama 3.2 3B Instruct outperforms Gemma 2 2.6B and Phi 3.5-mini on instruction following, summarization, prompt rewriting, and tool use. Tasks like extracting structured information from unstructured text, generating multi-step action plans, or maintaining coherence across a long tool-calling sequence all benefit noticeably from the additional parameter budget.
The context window is 128K tokens. The model supports multilingual text generation and tool calling, enabling agentic workflows. For example, it can process incoming messages, categorize them by urgency, and invoke calendar or notification tools accordingly.
The 3B is a natural step up from the 1B when you need more reliable instruction following. It's still substantially smaller than the 8B, which is the next tier up in the Llama 3.1 family.