Model loading was stopped due to insufficient system resources.
The message
Model loading was stopped due to insufficient system resources.What it means
LM Studio estimated how much memory the model needs at your settings, compared it with what your machine has free, and refused before loading anything. Nothing crashed.
What to do
Lower the context length or GPU offload and check with lms load --estimate-only. In the app you can click Load anyway; from the CLI or API you can't override it.
This is LM Studio's safety check refusing a load. From the command line it reads like this (the number depends on the model and your settings):
Error: Model loading was stopped due to insufficient system resources. Under the current settings, this model requires approximately 22.92 GB of memory, and continuing to load it would likely overload your system and cause it to freeze. If you think this is incorrect, you can adjust the model loading guardrails in settings.LM Studio calls these checks model loading guardrails. Before loading, it estimates how much memory the model will take and compares that with what's free. The app is closed source, so what follows comes from the docs and from LM Studio's bug tracker as of September 25, 2026.
What "under the current settings" means
The estimate covers the weights plus the context you asked for, so the same model can pass at 8,000 tokens of context and fail at 100,000. You can see the number without trying a load. The lms docs describe --estimate-only as printing "a resource (memory) estimate" and exiting, so add the context length you plan to use:
lms load --estimate-only qwen3.5-9b --context-length 32000Lowering context is the cheapest fix. Smaller quantizations of the model are the next one.
When the estimate is wrong
It can be too high. In #1631 LM Studio wanted 22.92 GB for a model the reporter ran in about 11.5 GB with plain llama.cpp, and will-lms (who contributes to LM Studio's lms CLI) replied on March 10 that "the model guardrails calculation is off for Qwen 3.5 models". Machines where the GPU borrows system RAM are another weak spot. In #1471 a Ryzen AI Max+ 395 with 96 GiB set aside as video memory was blocked, and the reporter found the check was counting only the roughly 22 GiB of system RAM left over.
Overriding it in the app, the CLI and the API
In the app, the dialog offers Load anyway, and the guardrail setting itself lives in LM Studio's settings (the error links you there). The CLI and the local REST API have no such button. The same March 10 reply confirmed "there is no way to override or change the guardrails setting headlessly", which is why a load that works in the window can fail from a script.
Editing the file directly isn't reliable either. A Mac user in #2017 set "mode": "off" and "alwaysAllowLoadAnyway": true in the model-loading guardrails section of ~/.lmstudio/settings.json and was still blocked. After a full machine restart lms load worked, but the API still refused. Their workaround, which we'd use too: load the model once with lms load, then point your app at the already-loaded model.
If the number is nonsense
An estimate of "approximately NaN undefined of memory" came from RWKV models in LM Studio 0.3.27 and was reported fixed in 0.3.28. If you see it on a current version, update and report it with the model name.
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: Model loading was stopped due to insufficient system resources.Under the current settings, this model requires approximatelycontinuing to load it would likely overload your system and cause it to freezeIf you think this is incorrect, you can adjust the model loading guardrails in settings.