Akshar is a free, open source dataset containing 3400 images of Gujarati language characters. Each folder in the dataset contains 100 images of a letter out of 34 characters. This dataset is relatively smaller in size than other datasets. It is intended for faster training of characters on any Neural Network model. 'Akshar' dataset is expected to be useful to test performance of various deep learning models. Gujarati alphabets are little more complex in shape than English alphabets. Thus, they are more challenging to classify using deep learning models. Dataset is free for anyone to use and improve by adding more font styles, complex letter orientations etc.