The image’s cost estimation depends on the method you use. If you use the processDocument method the Service bills the recognition task in A4 pages, and if you use the processFields method, the tasks will be billed as Fields.
I suppose you can try to use the following combination of methods: the submitImage method in order to add your images to one task, and then the processFields method to extract text from the fragments. At the moment one text field can contain small text snipped not longer than 200 characters (space characters are not included). If the field you recognize is longer, the extra characters will not be written into the output. Every 50 characters from the field are estimated as one Field. The limit with 200 characters is planned to expand in future, because sometimes the users need to recognize bigger fields. Unfortunately, I do nоt have information when exactly it will be changed.