Down Arrow Button Icon

Buying an expensive 5G-compatible phone to get on one of the mobile industry’s new superfast 5G networks offers a “mixed bag” in terms of actual download speeds, according to a new report.

Signals were sometimes difficult to find on some carriers’ networks while average download speeds occasionally failed to exceed current 4G LTE networks, according to phone speed testing firm Rootmetrics, which published the report on Wednesday.

“While 5G is clearly capable of offering much faster speeds than those of 4G LTE, it’s important to remember that we are still in the early days of 5G deployments, and it will take time for 5G to reach its full potential,” the firm said. “Our recent 5G testing revealed somewhat of a mixed bag in terms of speed results across the five cities we tested.”

Verizon, AT&T, and T-Mobile are spending tens of billions of dollars to add 5G. The technology offers downloads that are 10 to 100 times faster than an average 4G connection, opening the door to new applications like mobile virtual reality games and multi-angle live sports broadcasts.

But it will take several more years before the technology is widely available. Furthermore, it requires that users have 5G-compatible phones, which currently cost at least $1,000, though lower-priced models should be available soon.

At the end of 2019, Rootmetrics tested 5G networks in Indianapolis, Dallas, Chicago, Los Angeles, and Washington, D.C. It found that speeds in those cities varied widely, largely due the type of 5G technology that each carrier used.

Verizon 5G used high-frequency bands, known as millimeter wave, that carry vast amounts of data but don’t travel far. AT&T used lower bands, like 4G does, that travel greater distances but carry less data (this month, AT&T is adding millimeter wave service for consumers in some markets). Meanwhile, T-Mobile uses both approaches, but in the five cities reviewed by Rootmetrics, it mainly relied on lower bands.

Verizon had the fastest average 5G download speeds in most cities, but it had the smallest coverage areas, Rootmetrics found. In Los Angeles, for example, the company’s downloads averaged 247 megabits per second, while reaching a maximum speed of 627 Mbps, fast enough to download several TV show episodes in a few seconds. But its coverage was available in only 0.04% of the city. In Chicago, 5G Verizon’s downloads averaged 107 Mbps, and reached a maximum of 780 Mbps, with 3% coverage.

AT&T had somewhat broader coverage but slower downloads. In Los Angeles, for example, where coverage extended over 7% of the city, the company’s downloads averaged 36 Mbps—no faster than on 4G. Meanwhile, AT&T had a maximum speed in Los Angeles of 345 Mbps, more than double 4G’s maximum.

At the time of the test, AT&T lacked 5G in Chicago, but in Indianapolis, it averaged 65 Mbps, a little better than 4G, and a maximum of 386 Mbps, with almost 10% coverage.

T-Mobile’s 5G was slower and much more widely available. In Los Angeles, it averaged downloads of 23 Mbps—no faster than 4G—and topped out at 159 Mbps—slightly faster than 4G. But 5G was available across 39% of the city. In Chicago, 5G averaged 34 Mbps, with a maximum of 236 Mbps, with 44% coverage.

Sprint, which is being acquired by T-Mobile, is using a hybrid approach to 5G, which translated into a better balance between speed and coverage. In L.A., its 5G network averaged downloads of 81 Mbps, reaching a maximum of 201 Mbps, with 36% of the city covered. In Chicago, it averaged 137 Mbps and maxed at 250 Mbps with 16% coverage.

But T-Mobile is likely to use Sprint’s spectrum in its own 5G network after the $26 billion merger is completed.

The overall findings were consistent with an earlier report from testing firm Opensignal along with what Fortune found in recent tests.

Unlike Opensignal, which relies on download readings from apps on millions of consumer phones, Rootmetrics runs tests along specific routes and locations to compare carriers under identical circumstances. Rootmetrics’ testing may provide better comparisons between carriers but without the vastly larger number of download tests that crowdsourced data provides.

How Verizon, AT&T, and T-Mobile rank on 5G speeds

More must-read stories from Fortune: