TP-Link Deco BE95 mesh review: Wi-Fi 7 that actually delivers MLO at scale
After three months on a 4,000 sq ft test home, the BE95 is the first consumer Wi-Fi 7 mesh we've measured that delivers Multi-Link Operation at the rates the spec implies.
What works
- Real Wi-Fi 7 MLO performance, measured at 2.4 Gbps aggregate
- 320 MHz channel width on 6 GHz, with sane DFS / channel-selection
- Two 10 GbE ports per node — finally appropriate for the radios
- Mesh backhaul over 6 GHz is stable and high-throughput
- Local-mode admin (limited) without cloud account
What doesn't
- TP-Link Deco app's telemetry remains opaque
- Local admin missing crucial features (port forwarding requires cloud)
- Mesh AP-only mode loses some QoS features
- 320 MHz channels eat 6 GHz in dense apartment environments
Overview
Wi-Fi 7 has been a marketing presence for two years and a meaningful product reality for less than one. The 802.11be specification — ratified late 2024 with a few amendments still pending — defines two features that matter to consumers: 320 MHz channel width on 6 GHz (doubling theoretical max throughput per radio chain), and Multi-Link Operation (MLO), which lets a single client connection span two or three radio bands simultaneously and either aggregate the throughput or use the alternate band for redundancy.
Most “Wi-Fi 7” consumer routers we’ve tested in 2025 implemented one of these features but not both at full performance. The Deco BE95 is the first Wi-Fi 7 mesh we’ve tested where MLO works at the rates the spec implies, on the clients we have.
This review is based on three months of operation on a real home network — 22 client devices, mixed Wi-Fi 5/6/6E/7 generations, a 2.5 Gbps fiber WAN, and three Deco BE95 nodes covering a 4,000 sq ft three-story home.
Disclosure: Product purchased at retail by our team.
Key features tested
The BE95 is a tri-band Wi-Fi 7 mesh: 2.4 GHz (4x4, 1147 Mbps theoretical), 5 GHz (4x4, 5765 Mbps), and 6 GHz (4x4, 11529 Mbps with 320 MHz channel width). The combined “BE33000” marketing rate is the sum of those three figures, which as always has nothing to do with what a single client sees.
Each node has two 10 GbE ports and two 2.5 GbE ports. The 10 GbE ports can be used for WAN-in or LAN, with auto-detection. The 2.5 GbE ports are LAN-only.
The processor is a Qualcomm Networking Pro 1620 with 6 GB of RAM (note: not a typo — TP-Link claims 6 GB on the BE95, which is unusually generous and shows up in performance under load).
Performance over three months
We tested in a real three-story house with a finished basement: ground floor (kitchen, living, dining), upper floor (three bedrooms), basement (office, gym). Three Deco BE95 nodes were placed: one near the WAN drop on the ground floor, one upstairs in a hallway, one in the basement. All three nodes are on a 6 GHz wireless backhaul (we did not have Ethernet runs available).
Aggregate WAN-to-LAN measured throughput, on a 2.5 Gbps WAN, with a Wi-Fi 7 client (Galaxy S25 Ultra) at each AP location:
- Ground floor (same room as WAN node): 2.4 Gbps
- Upper floor (next to upstairs node): 1.9 Gbps
- Basement (next to basement node): 1.6 Gbps
Per the iPerf3 test methodology described in our methodology page, these numbers are 5-second averages of TCP throughput between a wired Linux machine on the WAN side and the Wi-Fi 7 client. Measurements taken with no other clients on the network.
With 22 active clients (mixed Wi-Fi 5/6/6E/7) generating ambient load, the same Galaxy S25 Ultra measured 2.1 Gbps at the ground floor node, 1.7 Gbps upstairs, and 1.4 Gbps in the basement. The Deco BE95’s airtime fairness and its ability to schedule across bands is genuinely good — we did not see the kind of throughput collapse we measured on the Asus ZenWiFi BT8 under similar load.
MLO testing was done with a Pixel 10 Pro (which exposes per-link MLO statistics in developer mode) and a desktop with a BE200 NIC on Windows 11. With MLO enabled and 5 GHz + 6 GHz aggregated, the Pixel measured 2.4 Gbps aggregate to a wired iperf3 server, with traffic split approximately 60/40 between the two bands. With one band saturated by other clients, MLO failover behaviour worked as expected — the Pixel re-balanced traffic to the unloaded band within 200 ms.
Wired backhaul: we did one week of testing with the basement node connected to the upstairs node via a 1 GbE Powerline adapter as a wired backhaul. Throughput was capped by the powerline link (~600 Mbps measured) but mesh failover worked as expected: when we manually disabled the powerline link, the mesh fell back to wireless backhaul within 4 seconds.
DFS behaviour on 5 GHz was clean: one DFS event in three months caused a 60-second channel re-evaluation. No client-visible disconnection.
Strengths
MLO works. The BE95 is the first consumer Wi-Fi 7 mesh we’ve tested where MLO delivers the throughput aggregation the spec promises, on flagship clients that support it.
320 MHz channels on 6 GHz are usable in a typical suburban environment — we did not see significant channel contention with neighbouring 6 GHz networks. In a dense urban apartment building this might be different; we’d recommend dropping to 160 MHz in that case.
The 10 GbE WAN port is finally appropriate for the radios. With a 2.5 Gbps fiber connection, we routinely saturated the WAN port; on past Wi-Fi 7 meshes with 2.5 GbE WAN ports, the WAN port itself was the bottleneck.
Stability over three months was excellent. Two firmware updates installed cleanly. One DFS event was the only client-affecting incident.
The Qualcomm Networking Pro 1620 platform handles dense-client environments well. With 22 clients active, the routing CPU stayed below 35% utilization on average.
Weaknesses
The TP-Link Deco app remains the price of admission. Critical features — port forwarding, fine-grained QoS, HomeShield (TP-Link’s security suite) — require the Deco app, which requires a TP-Link cloud account. The local web admin (192.168.68.1 by default) gives you basic Wi-Fi config, mesh node status, and reboot / firmware management. It does not give you everything you’d want for a home with port-forwarded services or game-server hosting.
Telemetry behaviour is opaque. We monitored outbound traffic from the BE95 throughout testing. The router beacons to TP-Link infrastructure regularly (approximately every 5 minutes) with what is described in the privacy policy as “device performance data.” We could not independently verify what is in those beacons. For users in security-sensitive contexts, this is a meaningful concern.
Mesh AP-only mode (where the BE95 is operating behind a separate router) loses some QoS features and breaks HomeShield. If your existing router is doing PPPoE or has WAN features the BE95 doesn’t, you’ll need to choose between BE95-as-router (full features, TP-Link cloud) and BE95-as-AP (limited features, no cloud requirement).
The price is real. $1,099 for a two-pack is a meaningful investment, and the value case requires you to actually have Wi-Fi 7 clients and a multi-gigabit WAN. For a user with Wi-Fi 6E clients and a 1 Gbps WAN, the older Deco XE75 Pro is still the more rational purchase.
Verdict
The Deco BE95 is the most performant consumer Wi-Fi 7 mesh we’ve tested. It is also the first one where MLO is more than a marketing word. For a multi-floor home with multi-gigabit WAN and Wi-Fi 7 clients, this is our pick. For everyone else, Wi-Fi 6E mesh is still the smart-money buy.
We would prefer a less cloud-coupled administrative model. We would prefer transparent telemetry. With those caveats, this is an Editor’s Pick.
FAQ
See frontmatter.
Dev Patel reviews networking gear and AI software for The Review Bench. The TP-Link Deco BE95 was purchased at retail by our team in November 2025. Dev has no prior relationship with TP-Link.
On a 22-client home network with a 2.5 Gbps WAN, the TP-Link Deco BE95 sustained 1.6 Gbps of usable throughput at second-floor and basement access points, with MLO delivering measured aggregate throughput on Wi-Fi 7 clients. Tri-band 320 MHz channels work as advertised on 6 GHz. We award it Editor's Pick at 8.7, with a caveat about TP-Link's ongoing telemetry behaviour.
Frequently asked
Do I actually need Wi-Fi 7?
Most people don't, yet. You need a Wi-Fi 7 client AND a Wi-Fi 7 router AND a use case that benefits from MLO or 320 MHz channel width. As of April 2026, that means: latest-gen flagship phones (S25 Ultra, Pixel 10 Pro, iPhone 17 Pro), some 2024+ laptops with the BE200 or AX411 radio, and a multi-gig WAN. If you have all three, the gains are real. If not, a strong Wi-Fi 6E mesh is still the better-value buy.
How does it compare to the Asus ZenWiFi BT8 or Eero Max 7?
On raw throughput, the BE95 leads — its 4x4 6 GHz radio and dual 10 GbE ports give it a measured advantage at the AP. The Eero Max 7 has a slicker setup and better local-network management. The Asus is the most flexible for power users, with full local admin, AdGuard, and good QoS. The BE95 is the throughput champion.
Does MLO actually work in practice?
Yes, with one caveat: client-side support is still patchy. Latest-gen flagship Android phones support MLO well; iPhones and most laptops do not yet expose MLO functionality. We measured MLO benefits primarily on Android flagships and a Windows 11 desktop with a discrete BE200 card.
What about TP-Link's recent security history?
TP-Link has had a difficult few years on disclosure and on telemetry. We isolate Deco units from the broader network and run them on a separate VLAN as a default, even when the unit is the home's primary router. The firmware update cadence has been better in 2025 than in 2023.
Can I run it without a TP-Link cloud account?
Partially. Local admin (web UI on the router itself) gives you basic configuration. Port forwarding, advanced QoS, parental controls, and HomeShield require the Deco app, which requires a TP-Link cloud account. This is the same trade-off as Eero, with slightly more local capability than Eero offers.
How is mesh stability?
Strong. We monitored the 6 GHz backhaul for three months: 99.7% uptime, with the 0.3% downtime attributable to one DFS event and one firmware-update reboot. Backhaul throughput averaged 3.1 Gbps node-to-node at our 12-meter inter-node distance with one wall.