Benchmarking 41 Open-Weight Models for Zero-Shot Intent Classification

A systematic evaluation of 41 models ranging from 135M to 9B parameters shows that instruction-tuned 3B models can outperform several 7B models in zero-shot intent classification. The study analyzes accuracy, confidence calibration, and deployment efficiency across eight English datasets.