That's not the model — the error toast at the bottom right tells the whole story: "OpenAI-compatible file content only supports image files for now; got 'text/other'."
When you attach a non-image file (your CourseNode.kt), LM Studio sends it through the OpenAI-compatible API as a "file content" part. The llama.cpp / llama-server engine only accepts image files in that part right now, so it silently drops the .kt text — the model receives an empty message, which is exactly why you see "This message contains no content."
Easy fix: don't attach the file — paste the code directly into the chat box as text. Then it goes through as normal text content and the model reads it fine. (File attachment for text documents would need an engine/path that supports it; on the OpenAI-compatible llama-server backend only images work at the moment — this is an LM Studio/engine limitation, not the model.)
Everything else in your setup is fine — Think + Code Interpreter are OK to leave on.
Tried the actual pasting of the code and it worked. I found out though that it alternates between throwing that error and behaving normally when I upload files. So when the error happens, I just wait a while and try again and it fixes itself
I tried it on Hermes though, and no issues there, so I'll continue using that
@octracus Good data point, thanks. The fact that it's intermittent and a quick wait-then-retry clears it points to LM Studio's file-handling timing on its side, not the weights — if it were the model it'd fail every time, not alternate. The Hermes difference comes down to how LM Studio routes attachments per model, not code-reading ability; both read pasted code fine.
So paste-as-text is the 100% reliable route, or run llama-server --jinja directly if you'd rather handle attachments outside LM Studio's UI. And no worries staying on Hermes if the upload flow fits your Kotlin work better — totally fair.
