Llama 3.2 3B Instruct
LegacyLlama 3.2 3B Instruct is a compact instruct model in Meta’s 3.2 generation aimed at mobile and edge scenarios with multilingual support on supported checkpoints.
Open weights LLM · Release — · Llama community license (see Meta)
Updated 1 day ago · Verified Apr 2026 · Score 78
Decision summary
Why teams reach for it, where it fits, and what to watch for — before you dive into specs.
Why teams choose it
- Ideal when privacy requires on-device inference.
- Combine with small rerankers for better retrieval QA.
Best use cases
- Use this when phone and tablet assistants
- Use this when offline field tools
Tradeoffs
- Narrower capability than 8B+ tiers.
- Quantization affects quality—measure perplexity on your domain.
Technical details
Modalities, benchmarks, and release context.
Modalities
What goes in and what comes out.
- Inputs
- text
- Outputs
- text
- Capabilities
- edge, multilingual
Benchmarks snapshot
Structured JSON for reproducible comparisons.
No benchmark data yet — see comparisons for relative performance.
Family lineup
Explore other versions in this family after you have the headline on this model.
Continue exploring
A short set of comparisons, nearby models, and links to go deeper — without repeating the same paths.
Compare with
Related models
Learn & build
Tools and curated destinations (max four).