Llama 3.2 3B Instruct

Meta's Llama 3.2 3B Instruct is a 3-billion-parameter text model released September 2024 under the Llama 3 community license, designed primarily for edge and on-device deployment but widely available on hosted inference. It consistently shows up as the cheapest substantive option on major providers — sub-$0.10 per million tokens on several platforms as of early 2026. At 3B parameters, quality-sensitive generation tasks like coding or complex reasoning are off the table, but classification, short-form summarization, and content moderation routing perform acceptably. The 131K context window is retained, which is useful for classification over long inputs. If your workload is volume-heavy and quality-tolerant, the 3B is worth benchmarking against the 8B before committing to the higher cost tier. Further price compression is likely as provider competition increases.