What Is the Best Mobile Carrier in Japan? (2026 Measured Data)

Ask this question in a Tokyo electronics shop and you will be told NTT DoCoMo, because DoCoMo has been the answer since the 1990s. Ask the measurement companies and you get a different answer, and it is not the same answer twice.

On Opensignal's most recent Japanese report, DoCoMo finishes last of four on Consistent Quality, the composite score for whether ordinary tasks complete without stalling. On Ookla's, DoCoMo wins a category Opensignal awards to somebody else. Neither firm is wrong. They are measuring different things, and the gap between them is the most useful thing a visitor can understand about Japanese mobile networks.

Here is what the data says, who wins what, and the number the coverage maps do not show you.

The scoreboard, metric by metric

Opensignal's April 2026 report covers 1 January to 31 March 2026. Ookla's covers January to June 2026. Both build their results from crowdsourced measurements taken on ordinary users' handsets rather than from carrier-supplied data, and they do not agree with each other.

Two of the column headings need unpacking. Consistent Quality is the share of tests where the connection was good enough for the things people actually do, such as video calling and streaming, so it is the closest thing to "does my phone just work". Reliability Experience is Opensignal's 0 to 1000 composite covering whether a connection succeeds and stays up at all.

au took 10 outright category wins plus one shared, out of 18, in the Opensignal report. That is the most of any Japanese operator, and the fourth consecutive edition of the report in which au has finished top overall.

Why "DoCoMo is best" stopped being true

DoCoMo wins coverage and 5G speed and finishes fourth on Consistent Quality, the composite measure of whether ordinary tasks complete without stalling. ITmedia described the result as an inversion: first for coverage and download speed, below the others on delivered quality.

Two pieces of context stop that from being a hatchet job.

The first comes from Opensignal's own Japan country manager. Japanese scores sit high globally across the board, he pointed out, and DoCoMo's figures are extremely high by international comparison even where they rank fourth at home. He attributed DoCoMo's position to spectrum constraints and base stations still under construction rather than to neglect.

The second is that DoCoMo says the same thing. Its CEO told reporters in August 2026 that quality has been improving but the company is "only halfway there".

DoCoMo's own account of what went wrong is specific. During the pandemic, traffic shifted to residential districts and the company redirected investment to follow it. When restrictions lifted in spring 2022, traffic in Shibuya hit 1.8 times its pre-pandemic level and the re-pointed network could not absorb it. The repair work is measurable: roughly 6,800 new 5G sites in FY2025, and a widening of the 800MHz LTE carrier from 20MHz to 30MHz once 3G shut down, which DoCoMo reports as about 1.5 times the indoor and underground capacity.

The benchmark this article leans on also has a public critic. SoftBank's CEO, asked about placing second to au, said his engineers consider Opensignal's evaluation criteria strange. He then added that losing is a fact, and that he would not hear excuses until SoftBank was first.

The number the coverage maps hide

Every Japanese carrier advertises population coverage of 99.9% or better. Each of them also files a quarterly coverage return with Japan's Ministry of Internal Affairs and Communications, the telecoms regulator, known as MIC. Those same filings report land-area coverage.

Those are the January to March 2025 filings. KDDI states it in its own words in an August 2025 press release: au's population coverage exceeds 99.9%, but because of Japan's terrain, land-area coverage is about 60%.

The gap is a definition, not a lie. Japan's population coverage metric works on a roughly 500-metre grid. If signal reaches more than half of a grid square, every resident of that square counts as covered. Coverage is scored on where people sleep rather than where they travel, and only outdoor base stations count. MIC publishes the caveat itself, noting that even at 100% population coverage there can be places where communication is impossible because of trees, mountains or buildings.

For a traveller this matters in one specific way. Japan's population lives in the coastal plains. Its hiking trails, ski areas and mountain roads are in the 27% to 40% of land nobody is counted in.

5G coverage, and the asterisk on SoftBank's lead

MIC reported national 5G population coverage of 98.6% as of 31 March 2026. Its own slide states that this is the four carriers' coverage overlaid, so no single network reaches it.

The most recent per-operator split is dated 31 March 2025, and it contains a reversal. The distinction that drives it is between two ways of building a 5G network. Sub6 means the mid-band spectrum below 6GHz (3.7GHz and 4.5GHz in Japan) that was allocated for 5G and delivers 5G speeds. The alternative is taking spectrum already used for 4G and re-pointing it at a 5G carrier, which lights up the 5G icon on your phone over a much wider area without delivering much extra speed.

SoftBank's chart-topping 97.1% rests on 84,653 base stations running re-farmed 4G spectrum flipped to a 5G carrier. On genuine Sub6 spectrum its coverage is the lowest of the big three. DoCoMo is the mirror image: lowest headline figure, highest real Sub6 coverage.

One line in that dataset settles a question travellers ask a lot. Millimetre-wave 28GHz coverage was 0.00% for DoCoMo, 0.69% for KDDI, 0.03% for SoftBank and 0.00% for Rakuten. Any marketing that invokes millimetre-wave gigabit speeds is describing something you will not encounter.

Rural Japan, the Alps, and the shinkansen

Three findings that decide whether your phone works away from the cities.

Low-band spectrum is where the incumbents separate from Rakuten. At MIC's March 2025 measurement, sub-1GHz population coverage stood at 99.74% for DoCoMo's 800MHz, 99.95% for KDDI's 800MHz and 99.94% for SoftBank's 900MHz. Rakuten's 700MHz band was at 4.33%. Low-band is the spectrum that bends around hills and penetrates buildings, so this is the single clearest predictor of whether a network works in a valley.

The Japanese Alps are the country's measurable weak spot. Ookla's prefecture data puts 5G availability at 35.2% in Osaka and 33.7% in Tokyo. In Yamanashi and Nagano, the two Alps prefectures, it falls to 9.1% and 9.8%. A separate nationwide survey by ICT総研, covering 729 measurement points across all 47 prefectures, found the slowest average downloads in Tochigi, Yamanashi and Nagano.

That survey also found no measurement point out of service for any of the four carriers, which is reassuring as far as it goes. Its points are stations, city halls, airports and landmarks, not trailheads and mountain huts.

4G is the thing to count on. Ookla's Q3 2025 data puts 4G availability above 97% for every major carrier, clustered tightly together. Whichever of the three networks your eSIM lands on, you will have 4G nearly all of the time, and 4G handles maps, translation, transit apps and video calls without complaint.

On the shinkansen, coverage of all tunnels on then-operating lines was finished in December 2020. The government's own announcement carried a caveat: signal is weak in places, and connections may drop depending on your carrier and handset. Real-world performance is still uneven. The engineering fix, which means mounting base stations on the trains themselves, is a research project with field trials scheduled for the second half of 2027, so expect interruptions through tunnels for now.

Where Rakuten Mobile sits

Rakuten is the interesting case, and the reason our Japan country plans do not use it.

Rakuten advertises 99.9% population coverage. Its own footnote states that the figure includes the partner line area, which is au roaming. Its last self-published own-network figures were 96% in February 2022 and 98% in October 2022, and it does not currently publish a headline own-network number.

That roaming arrangement is changing right now. KDDI's standing disclosure describes au roaming for Rakuten as a provisional measure running to the end of September 2026. In August 2026 KDDI's CEO said the company considers its role fulfilled and expects the contract to end on schedule.

The shape of the wind-down is counter-intuitive. What ends is downtown and residential coverage. Some rural areas are expected to continue under a narrower, time-limited arrangement. Rakuten confirmed in August 2026 that talks were still in their final stages, and the post-October map had not been published as of early September 2026.

None of this makes Rakuten a bad network. It won Ookla's Most Consistent Network and Best Gaming Experience awards for the first half of 2026 and took Opensignal's upload categories outright. It has also been fixing its best-known weakness: the shared base stations serving the Tokyo subway were widened from 5MHz to 20MHz, four times the capacity, with the Tokyo Metro and Toei rollout completed in July 2026. It is a network whose rural strength currently rests on an agreement being withdrawn, which is a different risk profile from the other three.

What we cannot tell you

Some travel sites claim that roaming eSIMs are deprioritised behind domestic subscribers on Japanese networks, and that this costs you 50% to 70% of your speed at busy times.

We looked for evidence and could not find any. No Japanese operator, no regulator and no independent measurement firm publishes a QoS priority policy for inbound roamers. The main source making the claim is a company selling Japan-local SIM cards in competition with roaming eSIMs, and it publishes no methodology, no measurement and no operator citation.

There is some counter-evidence. Third-party Speedtest results published by a Tokyo consumer site put foreign roaming eSIMs ahead of Japan-native tourist SIMs, which is not the ordering you would expect if roaming traffic were systematically throttled. It is also worth knowing that Japanese domestic tourist SIMs carry the same congestion-throttling language in their terms, so "buy a local SIM to avoid throttling" does not hold up either.

Priority tiering does exist in Japan as a paid product. KDDI markets a 5G Fast Lane feature, and ICT総研's survey methodology notes it had to disable that feature for fair testing. Whether operators apply lower priority to inbound roamers is undocumented in either direction, and we are not going to assert it.

What this means when you buy

Our Japan plans name their network on the plan card. That matters more than it sounds. At several data tiers the same price buys you a choice of carrier, and the post-cap throttle speed differs between them.

Prices are the referral price our own plan links carry, verified on 7 September 2026; list prices without a referral are higher. Plans marked Unlimited have a daily full-speed allowance and then continue at the reduced speed shown, so the throttle figure is what you live with once you pass the cap on a heavy day.

Reading the scoreboard back into a buying decision:

• Cities, and you want the fewest stalls. KDDI, which wins Consistent Quality and Reliability and has the least punishing throttle in our catalogue.
• You want raw speed in urban areas. SoftBank, which takes Opensignal's download category as well as Ookla's Best and Fastest Mobile Network awards.
• You want the fastest 5G, and your phone is a recent flagship. DoCoMo, which leads 5G download speed at 159.1 Mbps and wins Opensignal's Coverage award. Check the handset caveat below first.
• Rural, mountains or a long rail itinerary. See below, because the two coverage measures disagree and we are not going to pretend otherwise.

The verdict

au (KDDI) is the quality leader and SoftBank is the speed leader. Coverage is the contested one: DoCoMo takes the award, SoftBank reaches more of the land, and on the low-band spectrum that decides rural service all three are level. Those three networks hold every award between them across both measurement firms, and they are the three our Japan plans run on.

The more useful takeaway is the one the marketing does not carry: Japanese networks are strong enough that the differences between the big three matter less than the difference between being in a covered grid square and being outside one. Coverage of 99.9% of the population and 60% of the land is the same sentence, and which half of it applies to you depends entirely on where you are standing.