(Exit code: 18446744072635812000). Unknown error. Try a different model and/or config.
The message
(Exit code: 18446744072635812000). Unknown error. Try a different model and/or config.What it means
The engine process that loads the model (LM Studio's build of llama.cpp) crashed before loading finished. The long number is a Windows crash code mangled into an unsigned number, and it doesn't say which crash.
What to do
Open Developer, then Local Server logs, to see the real error. If it started after a runtime update, roll the runtime back; otherwise lower GPU offload and context length.
LM Studio shows this dialog when a model fails to load and the process doing the loading dies without a message:
🥲 Failed to load the model
Error loading model.
(Exit code: 18446744072635812000). Unknown error. Try a different model and/or config.The wording's been the same from LM Studio 0.3.37 in January to reports in August 2026, and 33 issues in LM Studio's bug tracker carry that number as of September 25, 2026. The app itself is closed source, so everything here comes from those reports.
What the number is
Most reports come from Windows, where crash codes are negative 32-bit numbers, usually written in hex starting 0xC0000. LM Studio reads the engine's exit code as an unsigned 64-bit number, which turns a small negative code into something just under 264, and then prints it through JavaScript, which can't hold that many digits exactly. We checked the arithmetic: 2,047 different Windows codes, everything from 0xC0000401 to 0xC0000BFF, all print as 18446744072635812000. So the number tells you the engine crashed hard. It can't tell you why (one commenter decoded it to a single code, but the rounding makes that impossible). A neighboring value, 18446744072635810000, turns up in other reports and is a different bucket of codes.
Where the real error is
The dialog drops whatever the engine printed. One reporter in #2234 found the actual CUDA failure "only appears in Developer > Local Server logs", not in the dialog. Look there first. The end of that log is llama.cpp's own output, so lines like CUDA error: out of memory or error loading model: mean what they mean in llama.cpp.
The two causes that keep coming up
A runtime update. LM Studio ships its engines as separate runtime packs (CUDA, Vulkan, ROCm, CPU builds of llama.cpp) that can update on their own. In #1370, Vulkan runtime v1.103.0 broke models that v1.102 loaded, and mattjcly, who contributes to LM Studio's lms CLI, replied on January 15 that a "candidate fix will be out shortly". In #2234 the CUDA 12 runtime auto-updated to v2.27.1 and a model that had worked for days stopped loading until the reporter went back to v2.25.2. If a model loaded yesterday and nothing else changed, this is the first thing to check.
Too much on the GPU. Another reporter got the same code "when trying to load Qwen3.5 with too high GPU offload", and in #2066 partial offload on an AMD card failed with this code while loading the whole model onto the GPU got further.
What to do
- Read the end of Developer, Local Server logs for the line before the crash.
- Open the runtime settings (Ctrl+Shift+R), tick "Include older versions", and select the previous version of your runtime. Turning off automatic runtime updates stops it happening again.
- Lower GPU offload and context length in the load settings, or switch to the CPU runtime once to see whether the model file itself is fine.
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.
Failed to load the modelError loading model.(Exit code: 18446744072635812000). Unknown error.Exit code: 18446744072635812000exit code 18446744072635810000