ESP-NOW Latency, Range and Packet Loss: What the Published Tests Actually Show

ESP-NOW Latency, Range and Packet Loss: What the Published Tests Actually Show

By Malik Hassan · Sources checked October 6, 2026

The ESP-NOW tutorial on this site shows you how to send a packet. It doesn't tell you how far the packet will travel, how often it will get lost, or what happens when you add more boards. Those answers exist in published tests, scattered across a conference PDF, two vendor-portal posts and a couple of abstracts. This is an attempt to put them in one place with the caveats attached.

The short answer

• Latency: In a 2026 lab test with two boards 2 m apart, ESP-NOW latency was about 2.4 ms, with no lost packets across 500 samples at each of three send rates. sadio
• Range: On ESP32-C6 boards with PCB antennas, packet success stayed near 100% to 150 m in an open field and fell to about 60% by 300 m. In forest it dropped below 50% by 125 m. espressif
• Long-range mode: ESP-NOW-LR stayed near 100% to 450 m in the open and still delivered 40% at 900 m, with throughput down to roughly 10 kbps. espressif
• Indoors: Only about half of eight nodes in a university building were reliable, and walls and open space explained the spread. espressif
• More senders: One exoskeleton study saw packet loss reach 8.6% as it went from one to four nodes. ups
• ESP-NOW vs Wi-Fi: One lab test put Wi-Fi latency about 40 times higher than ESP-NOW at 100 Hz, but a second study measured Wi-Fi at 20 to 40 ms, and neither reports its power-save setting. Don't quote the ratio. sadioespressif

How this review was done

The inclusion rule was simple: published tests since 2024 that measured ESP-NOW on real ESP32-family hardware and reported latency, loss, throughput or range. Five met it. Three were read in full (the 2026 JAIIO paper and two Espressif Developer Portal posts by University of Žilina authors). Two, marked †, are based only on their published abstracts. Espressif's ESP-IDF v6.1 documentation supplies the protocol facts. No new measurements were taken. The set is small and uneven, with different boards, distances and timing methods and no pooled statistics, so read it as a map of the evidence, not a benchmark.

The evidence at a glance

Study and setup	Measured	Headline result	Read with care
Montiel Torres, Balich & Balich, 55 JAIIO, 2026. Two ESP32-WROOM-32D boards, 2 m apart, line of sight, same channel as the Wi-Fi AP, 32-byte echo, 500 samples per rate.	Latency, jitter, loss at 10/50/100 Hz	ESP-NOW 2.35 / 2.57 / 2.41 ms, 0% loss. Wi-Fi (UDP) 113 / 107 / 105 ms, 1.4–2.8% loss.	Authors call it preliminary and assume symmetric round trips. The abstract says about 2.4 ms but the conclusion says under 2 ms. Wi-Fi power-save mode isn't stated.
Gancarčík, Šumský & Kubaščík, Espressif Developer Portal, Dec 2024. ESP32-C6-DevKitM-1 pairs with PCB antennas, open field and forest.	Latency, throughput, packet success by distance	ESP-NOW about 100% to 150 m open, 60% at 300 m, under 50% at 125 m in forest. LR about 100% to 450 m open, 40% at 900 m.	Espressif says it hasn't verified community posts. One board type, results as charts, and the one-way timing method isn't described.
Kubaščík, Espressif Developer Portal, Dec 2024. Eight ESP32 nodes sending every 250 ms to an S3/C6 gateway across a department, 30-second windows.	Packet success per node	About half the nodes reliable (best: nodes 1, 2, 4, 6). Loss tracked walls and open space.	Short windows, no repeats, results in charts.
Minchala Avila et al., IEEE ETCM 2024.† Jetson Nano controller with one to four ESP-NOW nodes in broadcast mode, compared with UDP/IP.	Latency, loss	Lower latency than UDP. Loss grew with node count, up to 8.6%.	Abstract only. Broadcast mode only.
Universitas Gadjah Mada thesis, Indonesia (year not shown).† Several transmitting boards sending to one receiver.	Latency, loss vs interval and sender count	Acceptable loss with four senders at a 5 ms interval and with up to 14 devices at 20 ms.	Abstract only. A thesis, and "acceptable" isn't defined. 55jaiio.sadio.org.ar +11

What the protocol itself promises

ESP-NOW is Espressif's connectionless Wi-Fi protocol. Your data rides inside a vendor-specific action frame, with no access point involved. The default bit rate is 1 Mbps. A v1.0 packet carries up to 250 bytes and a v2.0 packet up to 1,470. A v1.0 receiver truncates or drops v2.0 packets over 250 bytes, so mixed fleets need care. You can register 20 peers, of which up to 17 can be encrypted (the default is 7). espressifespressif

The line that matters most for reliability is easy to skim past. The send callback reports success when the frame is received at the MAC layer, which doesn't guarantee your application got it. Espressif suggests acknowledgements, retransmission and sequence numbers. espressif

Latency: a lab number and a field number are different numbers

The cleanest latency data is the JAIIO paper. Latency was half the round-trip time of a 32-byte echo, 500 samples each at 10, 50 and 100 Hz. ESP-NOW averaged 2.35, 2.57 and 2.41 ms, with jitter (standard deviation) near 1 to 1.5 ms and zero loss. Wi-Fi/UDP through an access point on the same channel averaged 105 to 113 ms, with jitter of 55 to 93 ms. One wrinkle: the abstract and table put ESP-NOW at about 2.4 ms, while the conclusion says under 2 ms. The table is the data, so 2.4 it is. 55jaiio.sadio.org.ar +2

The outdoor tests tell a different-looking story. ESP-NOW latency averaged under 20 ms out to 300 m in an open field, and about 30 ms at 150 m in forest. That isn't a contradiction. The boards differ, the distances differ, and the post doesn't describe how it timed one-way delivery, which is non-trivial because two ESP32 clocks share no time base unless you make them. Compare numbers inside a study, not across studies. espressif

Be suspicious of ESP-NOW-versus-Wi-Fi ratios

The JAIIO paper's Wi-Fi latency is far higher than the outdoor study's 20 to 40 ms. The JAIIO authors credit TCP/IP overhead and shared-medium access. Espressif's docs point at a second suspect: the default modem-sleep mode is the minimum power-saving setting, and while it's on, the delay in receiving Wi-Fi data can stretch to about one DTIM cycle. Calling esp_wifi_set_ps with WIFI_PS_NONE turns it off. Neither study says what its power-save setting was. From the text alone I can't tell which explanation is right, and I wouldn't quote a ratio without a rerun with power saving off. If you benchmark this yourself, log the power-save mode and the channel next to every result. In Arduino sketches, WiFi.setSleep(false) is the usual switch. espressif

Range: plan for 150 m, not 200

Repeated claims are cheap and measurements aren't. The indoor post's introduction cites a 200 m reach and a 250 kbps ceiling as background, yet the same faculty's outdoor post measured throughput up to about 400 kbps. The introduction's figures read like spec-sheet folklore, and the measured ones deserve more weight. espressifespressif

The field data, on boards with stock PCB antennas: in an open field, success stayed near 100% to 150 m and slid to about 60% at 300 m. In forest it fell below 50% by 125 m. Throughput peaked around 400 kbps up close and dropped to 50 kbps at 300 m. That peak is well under the 1 Mbps default bit rate, so expect application throughput to be a fraction of the nominal rate. espressifespressif

The long-range mode trades speed for reach. Success stayed near 100% to 450 m in open ground and fell to 40% at 900 m, while throughput slid from about 100 kbps at 150 m to about 10 kbps at 900 m. In forest, success dropped below 50% at 400 m. Nobody in this set tested external antennas, so treat these as figures for stock boards. espressif

Indoors, the building decides

Eight sensor nodes spread across a university department sent a packet every 250 ms to a gateway, with success counted over 30-second windows of 120 expected packets. About half the nodes proved reliable, and the authors tie the spread to walls and open space rather than distance alone. Their advice is to split a building into open-space segments, use more than one central receiver, and add software delivery control. With short windows and no repeats, I'd read this as a warning about geometry, not a number to design around. espressif

Adding senders

Two abstract-level sources speak to crowding. In the exoskeleton study, ESP-NOW broadcast had lower latency than UDP/IP, but loss grew as nodes went from one to four, peaking at 8.6%. The Indonesian thesis reports acceptable loss with four senders at 5 ms and with up to 14 devices at 20 ms. Without the full texts, take these as directional. The more transmitters share the channel, the more you should space and stagger them. upsugm

Six rules that follow from the evidence

1. Treat a success callback as a MAC-layer receipt only. Add your own acknowledgement, retry and sequence number. Espressif recommends this pattern. espressif
2. Plan for about 150 m in open air with stock PCB-antenna boards, and much less indoors. Walk-test before fixing node positions, logging the RX control data (RSSI lives there) from your receive callback.
3. Stagger your senders. Add a random offset to each node's schedule, and with a dozen nodes feeding one receiver, start at 20 ms spacing or slower and test downward.
4. Keep both callbacks short. They run from the high-priority Wi-Fi task, so queue the data and handle it elsewhere. espressif
5. Use v2.0 packets above 250 bytes only if every node supports them.
6. When you compare with Wi-Fi, state the power-save mode and channel.

What this review can't tell you

The set is small and the hardware mixed. Two sources are abstract-only. One study calls its own results preliminary, and Espressif says it hasn't independently verified the community posts. There's nothing here on ESP32-C5, on BLE or Thread coexistence, or on battery life. The JAIIO authors list encryption overhead as future work, so that's open too. 55jaiio.sadio.org.ar +2

Run the tests yourself

The most useful thing you can add to this evidence is a clean, repeatable test with the settings written down. Three are enough.

1. Latency: a 32-byte echo between two boards, with a sequence number and an esp_timer_get_time() timestamp in the payload. Take 500 samples each at 10, 50 and 100 Hz, with latency as round trip ÷ 2 and jitter as standard deviation. Run the Wi-Fi arm twice, with power save on and off.
2. Range: open ground in 25 m steps, four packets per second for 30 seconds per step (120 packets), logging success and RSSI. Repeat with the boards rotated 90°.
3. Crowding: 1, 2, 4 and 8 senders to one receiver at 100, 50 and 20 ms intervals, 60 seconds each. Count loss with per-sender sequence numbers, with and without random stagger.

Publish the CSVs, the code and a photo of the rig. Power-save settings are exactly the detail the studies above leave out.

FAQ

How far can ESP-NOW reach? In the one outdoor test reviewed (ESP32-C6 boards, PCB antennas), success was near 100% to 150 m in open air and about 60% at 300 m. Forest cut it below 50% at 125 m. Treat 150 m as a planning ceiling and test your own site. espressif

How fast is ESP-NOW in practice? Espressif lists a 1 Mbps default bit rate, but the outdoor test measured at most about 400 kbps, falling with distance. espressifespressif

Is ESP-NOW lower latency than Wi-Fi? In the 2 m lab test, yes: about 2.4 ms against 105 to 113 ms. The outdoor study measured Wi-Fi at 20 to 40 ms. The Wi-Fi figure looks sensitive to setup, so compare carefully. sadioespressif

How many devices can talk to one board? Espressif's docs allow 20 registered peers, up to 17 of them encrypted. In the abstracts reviewed, loss rose as senders were added, so pace your transmissions. espressif

Is delivery guaranteed? No. The send callback only reports MAC-layer success, and Espressif recommends your own acknowledgement and retransmission. espressif

References

• Montiel Torres, M. N., Balich, N. A., Balich, F. A. "Esp-now vs. wifi: an analysis of temporal determinism in high-frequency industrial telemetry." 55 JAIIO, 2026. PDF
• Gancarčík, S., Šumský, J., Kubaščík, M. "Exploring wireless communication protocols on ESP32 platform for outdoor applications." Espressif Developer Portal, 20 Dec 2024. Link
• Kubaščík, M. "Data transmission reliability over ESP-NOW protocol in indoor environment." Espressif Developer Portal, 19 Dec 2024. Link
• † Minchala Avila, L. I., et al. "Comparative Study of Wireless Communication Protocols for Lower-Limb Exoskeleton Systems." IEEE ETCM 2024. Abstract
• † "Unjuk Kerja Protokol ESP-NOW untuk Komunikasi Data Nirkabel Single Slave Multiple Master." Universitas Gadjah Mada thesis repository. Abstract
• Espressif Systems. ESP-NOW, ESP-IDF Programming Guide v6.1 and Wi-Fi Performance and Power Save