Phi-3 Mini 128K

Phi-3 Mini 128K is a 3.8B-parameter instruction-tuned model from Microsoft, released in April 2024. It was trained on a curated corpus of textbook-quality synthetic data, which lets it outperform several 7B-class models on reasoning and QA benchmarks despite its smaller footprint — a meaningful cost advantage when you are paying per token. The 131K context window is genuinely unusual for a sub-4B model and makes it viable for document extraction and classification workloads that would otherwise require a larger host. It ships under the MIT license, so there are no commercial restrictions. At this size, latency and hosting cost are the primary draws; manage expectations on complex multi-step reasoning.