Could not load text model:
The message
Could not load text model:What it means
KoboldCpp tried to load your model file, the loading code reported failure, and KoboldCpp quit. This line only names the file. The reason is a few lines higher in the console.
What to do
Start KoboldCpp from a terminal so the console stays open, and read the last error lines before this one. Update to the newest release if it says unknown model architecture, and lower GPU layers or context if it says out of memory.
You pick a GGUF file in the KoboldCpp launcher, press Launch, and a box titled Error pops up with (or the console ends on):
Load Text Model OK: False
Could not load text model: D:\kobold\gemma-3-27b-it-abliterated.q4_k_m.ggufThe path is from issue #2003. We read the code that prints it in KoboldCpp v1.122, released September 26, 2026. It's the last step of a sequence: print "Loading Text Model", hand the file to the loader, print whether that worked, and if it didn't, show this message and quit with exit code 3. There's no reason in it because KoboldCpp doesn't pass one along. The loader has already printed its complaint further up.
Getting the real error on screen
If you started KoboldCpp by double-clicking, the window can close before you've read anything. Run it from a terminal instead (Command Prompt or PowerShell on Windows, pointing at the .exe) so the output stays put. Then scroll up from "Could not load text model" to the first line that reads like an error. That line is what to search for and what to paste into an issue.
There's also a warning earlier on to look for: WARNING: Selected Text Model does not seem to be a GGUF file!. KoboldCpp prints it when the file name doesn't end in .gguf or .bin, which usually means the wrong file was picked.
What the maintainers have found behind it
A model newer than your KoboldCpp. The log says unknown model architecture: and a name in quotes. In #2003 that was 'qwen3-tts', and a collaborator answered that support was coming in the next release and "The message is accurate." New model families need a KoboldCpp built after llama.cpp learned them.
Pieces that don't belong together. The first problem in #2003 was a Gemma model loaded with a Qwen vision file. The collaborator's answer: "If you use gemma you must use its accompanying mmproj for vision support, mixing architectures is impossible." An mmproj is the extra file that gives a model image input, and it has to come from the same model family.
A damaged or partial download. For #1211, two months-old models that wouldn't load, the advice was to update and "make sure the files downloaded correctly." Compare the file size with the one listed on the download page.
Not enough GPU memory. In #1882 the owner pointed at cudaMalloc failed: out of memory in the log, which came up while it was reserving memory for the context. (That was a two-GPU machine with memory to spare, and the owner traced it to how the load was split across the cards, then changed that default in a later release. On one card it usually just means the model and context don't fit.) Lower --gpulayers (the default -1 lets KoboldCpp fit layers automatically, and 0 keeps everything off the GPU), shrink --contextsize, or try --usecpu once to see whether the file loads at all.
If none of those fit
Try a small, well-known GGUF on the same settings. If that loads, the problem is the specific file. If it fails too, the backend is the suspect (the owner's first question in one Mac report was whether KoboldCpp had been built with Metal support), so switch between CUDA, Vulkan and CPU and see which one gets through.
Other lines the same feature prints
Match yours against these if the one at the top of the page is not quite it. They come from the same code and mean related things.
Error: Could not load text modelCould not load text modelLoad Text Model OK: FalseWARNING: Selected Text Model does not seem to be a GGUF file! Are you sure you picked the right file?