Zigbee in 2026
Still the biggest installed base in the smart home, still limited by things that are not in the marketing.
What Zigbee actually is#
Zigbee is a full stack, not just a radio. The bottom of it is IEEE 802.15.4: 16 channels numbered 11 to 26 in the 2.4 GHz band, 250 kbit/s, O-QPSK with direct sequence spread spectrum. On top of that Zigbee adds its own network layer (routing, addressing, the Trust Center security model) and its own application layer (clusters, attributes, commands).
That matters because Thread uses the identical 802.15.4 radio and replaces everything above it with IPv6. Same physics, different plumbing. Thread vs Zigbee works through what that changes in practice; the short version is that a Zigbee device speaks Zigbee to a coordinator, and a Thread device is an IP host on your network.
Zigbee is governed by the Connectivity Standards Alliance, which was the Zigbee Alliance until 2021 and which also owns Matter. The same organization runs both, which is why the "Matter killed Zigbee" framing never made sense.
Three roles, and only three#
Every device on a Zigbee network is one of these:
| Role | Power | Sleeps | Routes for others | How many |
|---|---|---|---|---|
| Coordinator | Mains | No | Yes | Exactly one per network |
| Router | Mains | No | Yes | As many as you like |
| End device | Battery | Yes | No | Bounded by parent capacity |
The coordinator forms the network. It picks the channel and the PAN ID, and it acts as the Trust Center that hands out and manages keys. It is the one device you cannot lose without rebuilding: there is no second coordinator, no failover, and moving to new coordinator hardware generally means re-pairing everything unless your software supports a backup and restore of the network key and device list.
Routers are mains-powered devices that stay awake, relay traffic, and hold child tables. Bulbs, plugs, in-wall switches and mains-powered relays are almost always routers. They are free mesh infrastructure, which is the strongest argument for choosing Zigbee bulbs and plugs over Wi-Fi ones.
End devices are battery powered and sleepy. They attach to exactly one parent, sleep with the receiver off, and poll their parent for anything queued. That is what makes multi-year coin cell life possible, and it is why a NoAck in your logs against a battery device is usually normal rather than a fault.
The limit that actually bites: child capacity#
You will read that a Zigbee network supports 65,540 devices. Ignore that number. It is an addressing ceiling and has nothing to do with what your hardware will do.
The real constraint is how many direct children the coordinator and each router can hold. Old CC2531 coordinators managed around 20 direct children. Modern Silicon Labs EFR32MG21 and EFR32MG24 coordinators are far higher, and EFR32MG24 is displacing MG21 as the standard silicon in current dongles. Vendors rarely publish the exact figure per firmware build, so treat it as a design constraint rather than a number to plan against.
The design rule that follows is simple and it works: add mains-powered routers instead of hanging everything off the coordinator. A dozen Zigbee plugs and bulbs spread through the house give your battery sensors somewhere else to attach, which relieves the coordinator's child table and shortens every route.
Binding: the feature that makes Zigbee feel fast#
Binding creates a direct device-to-device link at the application layer. A Zigbee wall remote can be bound straight to a group of bulbs, so pressing the button sends a command from the remote to the bulbs without the coordinator being involved at all.
Two consequences follow, and both are real advantages:
- Lower latency. No round trip to the hub, no automation engine, no rule evaluation.
- It survives the hub being down. A bound switch keeps working while your coordinator reboots, your Home Assistant updates, or your internet is out. That property is why bound Zigbee scores well in what still works when the internet goes down.
The cost is that bound behavior lives in the devices, not in your automation platform, so it does not show up in your rules and it can surprise you later. Document it. Automations that people do not turn off treats bindings as the fail-safe layer under the software layer, which is the right way to think about them.
Green Power: the switches with no battery#
Zigbee Green Power covers energy-harvesting devices, typically kinetic wall switches that generate their own power from the press. They are genuinely useful for renters and for rooms with no wiring, and they behave unlike anything else on the network:
- They need a proxy to relay their frames, commonly a mains-powered bulb or plug that supports Green Power.
- They do not support binding.
- They do not appear in network scans, so your map will not show them and they will look missing when you audit the mesh.
None of that is a fault. It is just a different device class, and it is worth knowing before you spend an evening trying to find a switch that is working perfectly.
Zigbee 4.0, announced November 2025#
The CSA announced Zigbee 4.0 on 18 November 2025. It is fully backward compatible with Zigbee 3.0, which the alliance frames as protecting the more than a billion Zigbee devices already deployed worldwide. What it adds:
| Feature | What it does |
|---|---|
| Dynamic Link Key | Per-device key management rather than a static install code path |
| Device Interview | The network questions a joining device about its capabilities before admitting it |
| Restricted Mode | Constrains what a joining or untrusted device is allowed to do |
| Zigbee Direct | Onboarding over Bluetooth LE, so a phone can commission a Zigbee device |
| Batch Commissioning | Adding many devices in one operation instead of one at a time |
| Coordinated Sample Listening | Sleepy device to sleepy device communication |
| New sub-GHz PHYs | European 800 MHz and North American 900 MHz bands |
Two of those change the shape of Zigbee rather than just improving it. Zigbee Direct removes the worst part of Zigbee onboarding, which is putting a coordinator into permit-join and hoping. Sleepy-to-sleepy via Coordinated Sample Listening breaks the old rule that a battery device can only ever talk to its mains-powered parent.
The sub-GHz work is related to Suzi, the CSA brand for standards-based long-range sub-GHz mesh built on the Zigbee network layer. A certification program was planned for the first half of 2026; the actual opening date is not something to rely on yet. For now, if you need long range at sub-GHz in the US, Z-Wave Long Range is the shipping answer.
The channel problem#
Zigbee's 16 channels sit inside the same 2.4 GHz band as your Wi-Fi. A busy Wi-Fi network on the same frequencies will make a Zigbee network look broken: devices drop, commands take seconds, and Zigbee2MQTT logs MAC_CHANNEL_ACCESS_FAILURE, which means the radio could not get a clear moment to transmit. It is a spectrum problem, not a faulty dongle.
Zigbee2MQTT's guidance is to use one of the ZLL channels (11, 15, 20 or 25) for compatibility reasons; Home Assistant's ZHA docs point at 15, 20 or 25 for the same reason and otherwise recommend leaving the default alone. Both suggestions are about device compatibility as much as spectrum, and neither is a substitute for looking at what your own Wi-Fi is doing. The arithmetic behind which of those channels is genuinely clear, and which only looks clear, is in 2.4 GHz interference and channel planning. Feed your Wi-Fi channel into the channel planner to get the answer for your house.
The other aggressor nobody expects is a USB 3.0 port. USB 3.0 signaling produces broadband noise across 2.4 to 2.5 GHz, and a coordinator dongle plugged directly into a USB 3.0 socket sits inside it. The fix is a 1 to 2 meter USB 2.0 extension cable, which is exactly why Nabu Casa ships one with the Connect ZBT-2.
Coordinator hardware in 2026#
| Coordinator | Silicon | Notes |
|---|---|---|
| Home Assistant Connect ZBT-2 | EFR32MG24 | $49, detachable antenna, runs Zigbee or Thread, not both at once |
| Sonoff ZBDongle-E | EFR32MG21 | Widely deployed, inexpensive |
| Sonoff ZBDongle Plus / Max | EFR32MG24 | Current generation |
| ConBee III | Silicon Labs | Long-standing deconz and Zigbee2MQTT support |
| SMLIGHT USB and PoE adapters | Silicon Labs | PoE variants useful for placement away from the rack |
The single-radio limitation on combination dongles catches people out. If you want a Zigbee network and a Thread network in the same house, budget for two radios. And prefer a USB-attached radio to a network-attached one where you can: the host-to-radio protocol assumes a fast, lossless link, and putting Wi-Fi or PoE in the middle of it makes the whole stack less predictable. That reasoning is spelled out in Thread troubleshooting, and it applies identically to Zigbee.
Where Zigbee sits against everything else#
Against Z-Wave: Zigbee has more devices and lower prices, Z-Wave has a quieter band and better wall penetration. The full comparison is Zigbee vs Z-Wave.
Against Thread: same radio, different stack. Thread is IP-native and is what Matter uses. Zigbee needs a bridge to reach a Matter fabric, and going through one costs you features, as Matter bridges sets out.
Against Wi-Fi: no contest for battery devices, and no contest the other way for video.
Is Zigbee dead now that Matter and Thread exist?
No. The CSA owns both Zigbee and Matter and announced Zigbee 4.0 in November 2025, explicitly backward compatible with 3.0 to protect over a billion deployed devices. Zigbee still has the widest and cheapest device catalog in the smart home. What has changed is that new Matter-native devices tend to arrive on Thread or Wi-Fi rather than Zigbee.
How many devices can a Zigbee network really handle?
The 65,540 figure is an addressing limit and not useful. The practical ceiling is how many direct children your coordinator and each router can hold, which depends on the silicon and firmware and is rarely published. Old CC2531 coordinators managed roughly 20 direct children; modern EFR32MG21 and MG24 coordinators are much higher. Add mains-powered routers rather than testing the limit.
Does Zigbee work with Matter?
Only through a bridge. A hub such as SmartThings, Home Assistant, Hubitat or a Hue bridge can expose its Zigbee devices to a Matter fabric, but the bridge decides what gets through. Effects, scenes, firmware updates and fine-grained entities are commonly lost in translation. There is no such thing as a Matter over Zigbee device.
What does MAC_CHANNEL_ACCESS_FAILURE mean?
It means the Zigbee radio tried to transmit and never found the channel quiet enough to do so. It is congestion, not hardware failure. The usual causes are Wi-Fi on overlapping frequencies, a microwave oven running, or a coordinator plugged into a USB 3.0 port. Move the coordinator onto a USB 2.0 extension cable first, then look at your channel plan.
Should I use Zigbee binding or automations in my hub?
Use both, deliberately. Bind switches to the lights they control so the basics keep working with lower latency and survive the hub restarting. Use hub automations for anything conditional, scheduled or cross-device. The trap is forgetting a binding exists and then debugging a light that responds to a switch your automation platform knows nothing about.
Why do my Zigbee battery sensors show NoAck errors?
Because they are sleeping. A Zigbee end device turns its receiver off between polls, so a message sent to it while it sleeps will not be acknowledged immediately. Occasional NoAck entries against battery devices are normal. Persistent NoAck against mains-powered routers is not, and points at range, interference, or a device that has fallen off the mesh.
Do I need a separate Zigbee coordinator if my hub already has one?
No, and running two coordinators in one house is usually a mistake unless you intend two separate networks. If you already have a hub with a built-in Zigbee radio, use it. Add a USB coordinator only when you are moving to a different platform, when the built-in radio has poor placement, or when you deliberately want a second network on a different channel.
Primary sources
Specification and vendor documentation we checked while writing this page. Where a claim depends on firmware behaviour rather than a published spec, the page says so inline.