Why Cellular Battery Tests Disagree: The Modem Is the Variable Nobody Names

01 The test that started the argument

A three-phone drain test, an iPhone 18 Pro Max, a Pixel 11 Pro XL, and a Galaxy S26 Ultra, run on cellular by the channel In Depth Tech Reviews, drew a large audience and the usual crop of contradictory conclusions in the comments. Same phones, same video, opposite winners, depending on which viewer's carrier and room they mentally substituted. This article takes that dispute seriously: the disagreement is not noise around a truth, it is information about which variables the test design leaves uncontrolled.

The methodology is the standard one for the genre: phones set to comparable brightness, similar accounts and sync load, a mixed workload of streaming and browsing, run down over hours on cellular data, with percentages recorded over time. As consumer evidence goes this is genuinely useful, because cellular is the hardest case, the radio state most people actually live in on commute days, and the drain profile it produces is very different from the Wi-Fi desk tests most spec-sheet estimates assume.

What the format cannot do, structurally, is separate the variables it does not hold fixed: received signal strength at each phone's position, modem silicon and its power state machine, carrier aggregation choices the network makes per device, and the ratio of screen-on to standby time in the final tally. The winner it crowns is real for that room, that day, that tower. The question this article asks is how much of the result transfers to yours.

02 Signal strength: the hidden variable in every drain chart

Cellular transmit power is adaptive. When the tower is close and the received signal is strong, the phone can transmit at low power and decode with relaxed effort; when the signal degrades, the modem raises transmit power, increases retransmissions, and spends more time awake per delivered bit. Across the range from excellent to poor signal, the same workload can plausibly cost somewhere between 1.5 and 3 times the energy, a range consistent with published measurement studies rather than a single figure, because the multiplier depends on band, bandwidth, and how the network is loaded.

This is why identical phones in different rooms produce different drain charts. A test where one phone sat near a window on the 5G mid-band while another leaned on a marginal connection is not a phone comparison at all in the degraded minutes; it is a RF-position comparison wearing a phone comparison's clothes. Human reviewers cannot fully prevent this, two phones cannot occupy the same point in space, but placing both devices a fixed distance apart and logging per-phone RSRP, reference signal received power, would turn an accident into a documented covariate.

The chart below shows the estimated shape of the effect. The useful takeaway for readers is calibration: if a battery test was filmed in a strong-signal environment, its absolute numbers flatter every phone, and the phone with the weaker modem tolerance for poor signal will look relatively better than it would on your commute, where the multipliers compound differently.

03 Modem silicon: why the same SoC label hides different radios

Brand labels flatten the thing that matters. The three phones in the test do not share a modem: Qualcomm bundles its current integrated modem in the Snapdragon platform, Google ships its own Tensor modem design, and Apple fields a modem that has moved between vendor silicon and its own design across recent generations. Even two phones on nominally the same platform generation can differ in antenna tuning, power-amplifier efficiency, and the firmware that decides how aggressively to hold a weak signal versus handing off or dropping to a lower band.

Integrated designs changed the baseline. When the modem moved onto the SoC die, the penalty for shuttling data between separate chips largely vanished, and modern integrated modems idle and ramp with finer granularity. But integration does not equalize the state machines: one modem may hold a weak 5G connection at high transmit power for minutes where another falls back to 4G at a fraction of the cost, and that policy choice, invisible on any spec sheet, is worth more battery than most silicon differences.

The honest conclusion is narrower than comment sections suggest: a three-phone test measures the modem-plus-antenna-plus-firmware system in one RF environment. It does measure something real, arguably the most real thing in the phone, since radio is a top-three drain source for most users, but the result should be read as 'these devices, in this signal environment', not as a ranked ordering of radios in general.

04 Carrier aggregation and 5G standby behavior

Carrier aggregation, the practice of bonding multiple bands into one wider pipe, is a per-network decision that most reviews treat as invisible. It is not, for power. Bonded carriers mean more reference signals to track, more measurements of neighbor cells, and more frequent channel-state feedback; on mid-band 5G the overhead is modest, but in weak coverage the phone may keep measuring and reporting on carriers it can barely use. Two phones on the same carrier, same city, can be aggregating different band sets at different moments, which quietly changes the workload the radios perform.

5G standby is its own regime. Standalone 5G networks keep an always-on connection that is faster to wake but more expensive to hold; non-standalone modes lean on 4G anchors with different idle behavior. Estimating across published teardowns and measurement papers, idle drain differences between aggressive and conservative 5G standby policies plausibly reach 2 to 4 times in marginal coverage, which is why the overnight segment of any cellular test, where screen-on variables vanish, is the most information-dense part of the video, and the least discussed.

Smart 5G switching, the phone's habit of dropping to 4G when the speed does not justify the power, interacts with test design in a way that can flip conclusions. A phone that switches aggressively will post better standby numbers and worse speed-test numbers; a phone that holds 5G flatters the speed test and bleeds in the pocket. Neither policy is wrong. They are different answers to the same question, and a drain test that runs long enough to include sleep hours is mostly a test of which answer each phone chose.

05 Screen-on time versus standby: two different games

A display at fixed brightness costs an estimated 0.6 to 1.2 W on current flagships, which dwarfs every radio state except the worst transmit-power cases. In a screen-on segment, then, the phones are competing mostly on display efficiency and SoC decode cost, and the modem contributes a smaller share of the total. In standby, the display cost vanishes and the radios and background sync become nearly the whole story. The two segments are different games played by different subsystems, and blending them into one end-of-day percentage hides which phone won which game.

Splitting the ledger matters because users live differently. A heavy streamer on strong office Wi-Fi cares about the screen-on curve; a rideshare driver in marginal coverage cares almost entirely about the standby and navigation blend. Reporting the hours to empty as a single number answers a question almost nobody asked. The stronger format, already appearing in better channel tests, reports screen-on drain per hour and standby drain per hour as separate rows.

Re-reading the three-phone test through this lens: its most transferable findings are the relative shapes, how each phone's curve bends as the battery empties, whether any device shows a late-test cliff, and how much each loses overnight. Its least transferable artifact is the total-hours ranking printed in the thumbnail, which is a function of the room as much as of the radios.

06 How to read the next battery test you see

Before the ranking, look for the controls. Was brightness fixed and verified or set to auto? Same accounts and sync load? One phone or three on the same table, and if three, how far apart? Was the test Wi-Fi, cellular, or mixed, and if cellular, is signal strength logged anywhere? None of these questions has a bad answer; the sin is silence, because every uncontrolled variable is a degree of freedom the comment section will fight over.

Then read the curve, not the endpoint. A test that posts percentage readings every 30 to 60 minutes lets you see whether drain is linear, whether one phone burned early during setup and indexing, and whether the overnight segment costs 3% or 10%. Endpoint-only results, hours-to-empty with no intermediate readings, throw away most of the metrological value of the hours they took to film.

Finally, weight the environment honestly against your own. If the reviewer's signal was strong and yours is not, discount the phones with the least radio headroom; if the test is screen-on-heavy and you are a standby-heavy user, invert the emphasis. A drain test is a measurement of a phone-in-an-environment. The reader's job is to supply the environment term, and the reviewer's job, increasingly met by the better channels, is to log enough of it that the reader can.