What a local model does to your battery
Short answer
Sustained inference is one of the heaviest things a laptop does. Model size is the main lever: a 1.2 GB model costs a fraction of the power of a 6.6 GB one for the same question.
Why it costs more than it looks
Generation holds the GPU or CPU at high utilisation for the duration of the answer, which is exactly the workload mobile chips are designed to avoid sustaining.
It is bursty, though. A few seconds per answer and idle in between means a conversation costs far less than the peak draw suggests.
What to do about it
Run a small model on battery. Qwen 3.5 2B is 1.2 GB and scored 16/16 on our reading evaluation; the capability gap on document questions is smaller than the power gap.
Unload the model when you are not using the assistant. It stays resident to avoid a reload delay, and resident memory has a cost.
Ask fewer, better questions rather than many vague ones. Each answer is a burst of load.
An honest comparison
A cloud assistant costs you almost no battery, because somebody else's datacentre pays the power bill. That is a genuine advantage of hosted AI and it is worth stating plainly on a page like this.
Other ways people ask this
- local llm laptop battery
- Does running a local model drain my laptop battery badly?
Related
- Up a levelGuides: local AI, browser privacy and the hardware to run it
- Closely relatedRunning a local model on integrated graphics
- Closely relatedWhat a local model actually costs to run
- Closely relatedRunning a local LLM without a graphics card
- Closely relatedCan you run a local LLM on an NPU?
- Elsewhere on the siteCan your PC run a local AI model?
- Elsewhere on the siteHardware guide for local AI