I have observed that the understanding of drawing quality by the General LLM models differs for CAD exported drawings vs Scanned drawings.
CAD exported drawings capture models are able to capture the lines and curve objects. Were as Scanned PDFs it's just one image.
Is there any other open source model which reads scanned PDFs better than the generic models like gemini or claude opus ?
The constraint I see with any generic multi model LLMs os downsample to smaller size and model never sees the pixel.