From the open-source project: https://github.com/PDFMathTranslate/PDFMathTranslate-next The key advantage is that it preserves the original PDF layout.Supports application/json and multipart/form-data request bodies.Supported PDF formats: a valid PDF URL / base64 with header / direct upload (multipart/form-data). If the PDF size exceeds 20M, only the PDF URL method is supported.Pricing: tokens consumed by the translation LLM model + 0.001PTC * number of translated PDF pages