Llama 3.2 1B Instruct

Meta's Llama 3.2 1B Instruct is a 1-billion-parameter text model released September 2024 under the Llama 3 community license, the smallest variant in the Llama 3 family. Its primary design target is on-device and mobile inference — phones and edge hardware where the 3B or 8B models exceed memory budgets. Hosted inference is available across a handful of providers, typically at sub-$0.05 per million tokens, but the quality ceiling is low: at 1B parameters it is not competitive with the 3B on most quality-sensitive tasks including summarization and instruction-following. Useful as a proxy for latency testing at the smallest weight class, or for truly budget-constrained triage pipelines where a small fraction of acceptable responses is acceptable. Broader deployment as an on-device base layer, rather than a hosted API, is where this model is expected to see most usage going forward.