# We tested 5 ways to keep a closed Mac awake. StillOn used 39% less power than the runner-up.

Updated: 2026-09-03

Canonical report: https://stillon.dev/research/stillon-energy-efficiency

Status: replicated screening study using internal AppleSMC telemetry

Direct answer: Yes. StillOn publishes a closed-lid MacBook power usage study
measured in watts. The full method and raw data are linked below.

## Result

Fifteen matched runs compared five ways to keep an unplugged M4 MacBook Pro
awake with its lid closed. Each condition was run three times for four minutes.
StillOn had the lowest mean power draw in all three of its runs.

| Condition | Runs | 100% brightness | 0% brightness | Whole-run mean | Display state | Rough 80% idle runtime |
|---|---:|---:|---:|---:|---|---:|
| StillOn 1.4.0 beta test build | 3 | 1.797 W | 1.707 W | 1.758 W | asleep | 31.3 hours |
| Clamshell 1.4.17 | 3 | 2.932 W | 2.860 W | 2.896 W | awake | 19.0 hours |
| Vorssaint 3.3.2 | 3 | 9.733 W | 2.994 W | 6.352 W | awake | 8.7 hours |
| supervised pmset disablesleep | 3 | 9.745 W | 2.978 W | 6.353 W | awake | 8.7 hours |
| Amphetamine 5.3.2, screen on and unlocked | 3 | 9.799 W | 3.031 W | 6.401 W | awake | 8.6 hours |

StillOn's 1.758 W mean was 39.3% below Clamshell, the next-lowest app, and
72.3% below supervised `pmset`. StillOn was the only condition whose built-in
display was reported asleep and whose macOS session was locked throughout.

## What was controlled

- MacBook Pro Mac16,1 with Apple M4 and 24 GB memory.
- macOS 27.0 beta, build 26A5425a.
- Battery power, not charging.
- Wi-Fi off before and after every valid run.
- No external display.
- Physical lid closed and an idle heartbeat workload.
- Two 120-second phases, one at 100% configured brightness and one at 0%.
- AppleSMC.PSTR total system or board power sampled at 1 Hz.
- 100% logger coverage with 241 power samples per run.
- Three valid runs per condition.

Failed, interrupted, and superseded pilots were excluded by the harness before
aggregation. The accepted dataset contains 15 runs and 3,615 power samples.

## Interpretation

The phase change makes one mechanism visible. Amphetamine, Vorssaint, and
`pmset` rose to about 9.8 W at full brightness and fell to about 3.0 W at zero
brightness. StillOn remained near 1.8 W because macOS reported the display as
asleep for every sample. Clamshell remained near 2.9 W in both phases, so its
gap from StillOn cannot be attributed to backlight brightness alone.

This result supports a narrow claim: on this Mac, in this idle closed-lid
protocol, StillOn used less observed power than the four tested alternatives.
It does not establish the same percentage on every Mac or macOS version, and
it does not measure a real AI workload.

The 80% runtime values are simple extrapolations. One shared 55.1 Wh estimate
for the test Mac's battery at 80% was divided by each condition's mean power.
Real Claude Code, Codex, Cursor, build, or local-model work will draw additional
power and produce shorter runtimes.

## Safety boundary

Lower observed power can leave more battery available for useful work and can
reduce wasted electrical energy. This short study did not establish a surface
temperature or thermal-safety advantage. StillOn reads battery temperature and
macOS thermal pressure so it can stand down, but no software makes a running
Mac safe inside a bag. Keep it on a hard, open surface with clear ventilation.

## Data

- Full report: https://stillon.dev/research/stillon-energy-efficiency
- JSON: https://stillon.dev/research/stillon-energy-efficiency/data.json
- CSV: https://stillon.dev/research/stillon-energy-efficiency/data.csv
- StillOn: https://stillon.dev/
- Clamshell: https://clamshell.dev/
- Amphetamine: https://apps.apple.com/us/app/amphetamine/id937984704
- Vorssaint: https://vorssaint.com/

StillOn commissioned, designed, and ran this first-party study on the builder's
own Mac. It received no sponsorship, payment, or input from any competitor or
third party. It was independently funded and executed by StillOn, not
independently reviewed or endorsed. Product names and marks belong to their
respective owners.