Conow Atlas 6000 AC review (CME07-2BL600-08): what you get back depends on how warm it starts charging

Choose a home battery on a single efficiency figure and you are calculating with a number that does not exist for this one. How much of your power comes back from the Conow Atlas 6000 AC depends on how hard you run it, and at full power also on how warm the battery is when charging begins. It costs € 1,720.96. third partyElektramat.nl, asking price including VAT, in stock · 3 October 2026

At 3000 W we charged it twice from about 10 to 80% and let it run empty again. After the cool start, 86 of every 100 kWh that went in came back; after the warm start, 81. Both within what Conow promised beforehand, at least 81%, and still not the same. The difference is in the charging: halfway through the warm charge the charging power dropped from 3039 W to 2623 W, the charge took 102 instead of 92 minutes, and it needed 5.9% more power for the same charge. own measurementour techbench, inline kWh meter at the socket and the battery temperature from the registers; one cool and one warm charge · 3000 W setting · 20 September 2026 manufacturerConow, e-mail of 8 September 2026 · 8 September 2026

This review is a first look, on the firmware as delivered. According to Conow, new firmware arrives in mid-October; after that, a revisit with a new measurement follows. manufacturerConow, e-mails of 17, 18 and 28 September 2026 · 28 September 2026

Every figure says who says it, and every measurement the setting and the day. I do not give it a grade. After that, you decide.

Key takeaways

• What you get back depends on how warm the battery is when charging begins. Charge at 3000 W right after a heavy discharge and, at a reference price of 25 cents per kWh, you lose € 1.20 more per 100 kWh than with a cool start: € 4.68 against € 3.48. Halfway through the warm charge, at a battery temperature of 45.7 °C, the charging power dropped from 3039 W to 2698.6 W, and the last quarter ran at 2623 W. Discharging did not change with temperature. own measurementour techbench, inline kWh meter and the battery temperature from the registers; two charges and three discharges · 3000 W setting · 20 September 2026 derivedfrom the round trips 0.861 and 0.813 and a reference price of € 0.25 per kWh
• At 1500 W, 87 kWh of every 100 kWh you put in comes back. Measured socket to socket: a round trip of 0.87, with charging at 91.3% and discharging at 95.0%. Conow promised at least 81%. own measurementour techbench, inline kWh meter, one full discharge and one full charge · 3000 W setting · 18 September 2026
• It reaches its stated power, and a little more: 3061 W charging and 3006 W discharging, against 3 kW on the datasheet. Out of the box it is limited to 800 W on the grid. Everything in this review above 800 W was measured on the 3000 W setting, which you unlock in the app. own measurementour techbench, at the socket · 3000 W setting · 16 September 2026 manufacturerAtlas 6000 AC datasheet · 10 September 2026
• At low demand it still delivers, and at rest it takes nothing measurable from the grid. Ask for 30 W and 27.32 W arrives at the socket; the AEG SolarCube Pro on the same bench is a net importer below 41 W. That low power is not cheap: at 100 W a fifth to a quarter is lost in each direction, 0.754 charging and 0.788 discharging. At rest the grid side read 0.0 W for eight hours; what it uses then comes from the battery and is under 5 W. Conow promises less than 10 W with Wi-Fi on. own measurementour techbench, inline kWh meter · 3000 W setting · 20 September 2026 own measurementour techbench, inline kWh meter, eight hours idle · not recorded; idle · 19 September 2026 manufacturerConow, e-mail of 8 September 2026 · 8 September 2026
• If your consumption changes faster than about half a minute, it makes the result at the meter worse than doing nothing. Starting or reversing takes 12 to 35 seconds, adjusting while running 4 to 9. own measurementour techbench, step response over Modbus and through an emulated meter · 3000 W setting · 16 September 2026
• Local control works, but then you take over the whole control loop. The switch the app calls "Modbus TCP" turns off self-consumption and the AI modes, and it is RS-485, not TCP. If your own control stops sending while the connection stays open, it keeps charging; if the connection drops, it stops after about a minute. own measurementModbus RTU over RS-485; watchdog test and connection cut, at the socket · 3000 W setting · 22 September 2026
• Conow acknowledged seven errors in documentation and firmware and put part of the fixes on the mid-October firmware. That includes a register for the DC side, explicitly for measurements like these. They are written out below, one by one. manufacturerConow, e-mails of 16 to 28 September 2026 · 28 September 2026

How it was measured

• Unit: CDJCAC2605200086
• Firmware: Main V1.0.0 (1.0.8) / MCU V1.01.06, read by hand from the app (there is no firmware register); automatic updates were off, and the build was unchanged at the end of the campaign
• Test date: 2026-09-14 to 2026-09-21
• Test duration: about eight days on the bench, including a full charge and discharge cycle at 1500 and 3000 W, a ladder from 100 to 3000 W in both directions, a watchdog test and eight hours idle. On 22 September one more test followed with the connection cut, and on 3 October the backup port was put to use, not measured
• Grid setting: everything above 800 W on the 3000 W setting, which you unlock in the app; the setting is given with every figure in this piece. Out of the box it is limited to 800 W on the grid, and warm charging on that setting was not measured
• Instrument: our techbench. On the grid side a metering plug until 17 September, then a fixed inline kWh meter with MID approval. State of charge, battery voltage and battery temperature come from the device's registers, over RS-485; the cooling fins were measured with a thermal camera at about half a metre. The DC side cannot be read on this firmware
• Measurement log: conow-atlas-2026-09-14
• Method: the power ladder follows ladder-v1 of our test method (in Dutch); the full cycles, the idle measurement and the control tests are separate measurements, each with its date

One unit is one unit. Everything below is about this device on this firmware in these weeks, not about the model range and not about the brand. And the setup has a limit: without a register for the DC side, the bench only sees what happens at the socket. That is why the efficiency per step comes from the device's energy counter, and every step carries its error bar. On one step, 200 W charging, that counter reached 19 of the 20 counts we use as a floor; there is no figure there. own measurementour techbench, the device's energy counter; ladder-v1 and ladder-low-v2, 19 to 21 September, with an error bar per step · 3000 W setting · 21 September 2026

What it is

A plug-in battery of 6 kWh with a 3 kW inverter inside. It plugs into a socket, charges on your panels' surplus and feeds back into the house, controlled against a meter. One cabinet of 800 x 315 x 236 mm and 55.8 kg net, with LFP cells, IP65, a backup port, Wi-Fi for the app and an RS-485 port for Modbus. The datasheet gives 6028 Wh nominal and 3000 W charging and discharging, and for the backup port 3000 W continuous and 4500 VA peak for 3 seconds, with a switchover within 10 ms. In its e-mail, Conow states 6000 Wh usable. manufacturerAtlas 6000 AC datasheet, attached to the e-mail of 8 September 2026, without version or date · 10 September 2026 manufacturerConow, e-mail of 8 September 2026 · 8 September 2026

Out of the box it is limited to 800 W on the grid. The datasheet gives three settings for the grid port: 799, 800 and 3000 W. According to Conow, you unlock the full 3000 W yourself in the app, once it is connected to a dedicated circuit or approved by a qualified installer; you confirm that yourself in the app. manufacturerAtlas 6000 AC datasheet, AC output row; Tuya Smart on behalf of Conow, e-mail of 8 September 2026 · 8 September 2026

In the Netherlands it has to be registered with the grid operator in every setting. Because you unlock the 3000 W yourself in the app, the Atlas is a 3000 W unit for the grid operator, even when it is set to 800 W: a setting in an app is not decisive, only a factory limit without an app is. third partya spokesperson for Netbeheer Nederland, e-mail to Jay's Desk · 14 August 2026

The brand calls itself "powered by Tuya Smart", and Tuya Smart is listed on the New York Stock Exchange. Our contact for this review also ran through Tuya Smart. Its type number is CME07-2BL600-08. Who repairs the Atlas in the Netherlands when something breaks is not in what Conow sent us; that is among the open points. third partyConow press release of 12 March 2026, via stocktitan.net · 12 March 2026

What comes out

The round trip. Put 100 kWh in at 1500 W and you get 87 kWh back, socket to socket: a round trip of 0.87. Charging runs at 91.3%, discharging at 95.0%. A full charge, from 9 to 100%, took 6053.8 Wh from the socket and 4 hours and 4 minutes. Conow promised at least 81% beforehand, and at 1500 W it is well above that. own measurementour techbench, inline kWh meter; discharge on 17 September, charge on 18 September · 3000 W setting · 18 September 2026 manufacturerConow, e-mail of 8 September 2026 · 8 September 2026

Later Conow sent its own measurements over a full cycle, from empty to full: 87.22% at 800 W, 85.00% at 2000 W and 81.70% at 3000 W. The higher the power, the lower their round trip, and according to Conow our result falls within their expected range. Their cycle is not ours, so the figures stand next to each other, not against each other. manufacturerConow, e-mail of 22 September 2026 · 22 September 2026

Round tripwhat comes back of a kWh

The harder it works, the less comes back. Run it at full power and you get less into the house per kWh than when it runs gently. At 1500 W it converts 95.0% when discharging, at 3000 W 91.8%. Per step, at 3000 W it sits around 0.92. own measurementour techbench, inline kWh meter, full discharges at 1500 and 3000 W · 3000 W setting · 18 September 2026

And the gentler, too. At the low end it drops harder. If your house asks for 100 W at night, a fifth to a quarter is lost in each direction: at 100 W it converts 0.754 when charging and 0.788 when discharging. At 200 W discharging it is 0.852. Between 300 and 1000 W the steps lie between 0.893 and 0.969, but the error bars overlap there: every value is in the chart, and no peak or ordering can be read from it. From 1500 to 3000 W they lie between 0.980 and 0.921. own measurementour techbench, the device's energy counter; ladder-v1 and ladder-low-v2, 19 to 21 September, with an error bar per step · 3000 W setting · 21 September 2026

Efficiency per stepfrom 100 to 3000 W, charge and discharge, with error bars

Warm or cool charging. Charge it at 3000 W right after a heavy discharge and you get less back, and the charge takes longer. We charged it twice at 3000 W, from about 10 to 80%. The cool charge started with the battery at 25.1 °C, stayed below 39.8 °C and held just over 3000 W all the way. We made the warm one hot on purpose: first empty at 3000 W, then charging again shortly after, with the battery at 41.2 °C. At 47% state of charge, with the battery at 45.7 °C, it was still at 3039 W; one second later at 2698.6 W, and the last quarter of the charge ran at 2623 W. own measurementour techbench, inline kWh meter and the battery temperature from the registers; one cool and one warm charge · 3000 W setting · 20 September 2026

Charged cool and charged warmtwo charges at 3000 W, from about 10 to 80%

What do you notice? The cool charge took 92 minutes for 71% added, the warm one 102 minutes for 70%. The warm one needed 68.5 Wh per percent of charge, against 64.7 Wh for the cool one: 5.9% more power for the same charge. Over the whole round trip, 0.86 came back after the cool start and 0.81 after the warm one. own measurementour techbench, inline kWh meter, one cool and one warm charge · 3000 W setting · 20 September 2026

It is in the charging, not in the discharging. Three times we ran it empty at 3000 W, with the battery starting at 24.6, 36.3 and 35.9 °C, and all three times 55.7 Wh per percent came out. The discharges reached 46.7 °C and the power did not drop. own measurementour techbench, inline kWh meter, three full discharges · 3000 W setting · 20 September 2026

Conow wrote beforehand that it derates above 45 °C when charging and above 50 °C when discharging, and later by how much: when charging above 45 °C cell temperature to about 2590 W, above 50 °C to about 1940 W; when discharging above 50 °C to about 2590 W. Our warm charge dropped at 45.7 °C to 2623 W, close to what Conow states. The discharges stayed below 50 °C and did not drop. Conow explains why a warm charge costs more power with heat in the battery's internal resistance, and according to Conow it is within what they expect. The explanation is Conow's; we only measured the difference. manufacturerConow, e-mails of 8, 17 and 28 September 2026 · 28 September 2026

The room was cool, so the heat came from that discharge. Whether a warm spot, an attic or a cupboard, does the same, we did not measure. I expect it does, but that is among the open points.

What that is in euros. In every review we use the same reference price, 25 cents per kWh, so the reviews line up; it is not your tariff. Charge 100 kWh at 3000 W and you lose € 3.48 with a cool start and € 4.68 with a warm one: a difference of € 1.20 per 100 kWh. derivedfrom the round trips 0.861 and 0.813 and the reference price of € 0.25 per kWh

What a kWh from this cabinet earns you on your own contract, you can work out with the energy cost tool: enter your contract, your consumption and your feed-in, and it gives the price per kWh you take from the grid and what a kWh you feed back earns. The tool has no input for round-trip efficiency, so you do that step yourself. At 1500 W, count on 0.87 of what goes in; at 3000 W, on 0.86 if it starts charging cool and 0.81 if it starts warm. own measurementour techbench, inline kWh meter · 3000 W setting · 20 September 2026

How much comes out. From full down to the reserve, 5192.6 Wh came out at 1500 W, at the socket, from 100 to 10% state of charge. At 3000 W, 3.4% less came out of the same band. Conow states 6000 Wh usable, which is a different quantity: the bottom of our measurement is the reserve you set yourself, and we measure at the socket, after the inverter. That is why the two figures stand side by side here, and why one does not refute the other. own measurementour techbench, inline kWh meter, full discharges at 1500 and 3000 W · 3000 W setting · 17 September 2026 manufacturerConow, e-mail of 8 September 2026 · 8 September 2026

The power. Ask for full power and you get what the datasheet promises, and a little more. With 3400 W requested, 3006 W came out and 3061 W went in, at the socket; it gave no more than that. Conow gives a tolerance of 1% of the power plus 1% of 3000 W, a band of 2940 to 3060 W. The 3061 W when charging is 1 W above that band, right at the edge. own measurementour techbench, at the socket · 3000 W setting · 16 September 2026 manufacturerConow, e-mail of 28 September 2026 · 28 September 2026

Does it do what you ask. Yes, and on the cautious side when discharging: it never delivers more than you ask, so it does not push anything onto the grid you did not want. Across the whole ladder it stayed 2.5 to 11.4 W below the request at every step when discharging. Charging stayed within 2.3 W of the request up to 1500 W; from 2000 W up, 16 to 28 W more went in. own measurementour techbench, inline kWh meter, ladder-v1 in both directions · 3000 W setting · 19 September 2026

Requested against delivereddeviation per step, charge and discharge

At rest and at low demand

At low demand. If your house asks for only a few tens of watts, it still delivers them. At 30 W requested, 27.32 W arrived at the socket, and that gap stayed at about 2.4 W from 30 to 150 W requested. The AEG SolarCube Pro on the same bench is a net importer from the grid below 41 W. own measurementour techbench, inline kWh meter, five steps from 30 to 150 W · 3000 W setting · 20 September 2026 own measurementour techbench, AEG SolarCube Pro, metering plug and DC side · 3000 W setting · 12 September 2026

What it costs the battery, the bench cannot see: the DC side cannot be read on this firmware. If your house often asks for little, the fixed loss weighs most. Conow itself states 30 to 50 W of fixed loss, for cooling, control electronics and magnetics, and writes that efficiency improves at higher power. You see that in the steps: at 100 W, 0.788 arrives when discharging and 0.754 goes into the battery when charging, so a fifth to a quarter is lost in each direction. That is exactly the power a house asks for at night. manufacturerConow, e-mail of 28 September 2026 · 28 September 2026 own measurementour techbench, the device's energy counter; ladder-v1 and ladder-low-v2, 19 to 21 September, with an error bar per step · 3000 W setting · 21 September 2026

At rest. When it stands still, you do not see it on your meter. Over eight hours idle the grid side read 0.0 W, and the import and export counters did not move by 0.001 kWh. What it uses at rest comes from the battery and is under 5 W. The counter cannot do better than that. Conow promises less than 10 W at rest with Wi-Fi on; it meets that, but by what margin we will only know once we can read the draw from the battery. own measurementour techbench, inline kWh meter, eight hours idle · not recorded; idle · 19 September 2026 manufacturerConow, e-mail of 8 September 2026, data list item 4; not on the datasheet · 8 September 2026

Control and integrations

Controlling it yourself takes over the whole loop. Control it with your own script or an energy manager and you switch off self-consumption and the AI modes. The switch in the app is called "Modbus TCP", but it is a mode, not an extra feature: on means it waits for commands from outside, and you go back with the same switch. It is not TCP either. It is Modbus RTU over the RS-485 port, and you need an adapter for it. Over that port, control is local, without a cloud. This unit has no real Modbus TCP; Conow is aiming for the end of this year and called the label its own mistake. own measurementModbus RTU over RS-485 through a network bridge · n/a · 16 September 2026 manufacturerConow, e-mail of 17 September 2026 · 17 September 2026

Discharging needs two things at once. Send only a command and nothing happens when discharging. It discharges only when your command and the meter both show demand, and then it delivers the smaller of the two. Charging is the other way round: the command is enough there. Conow confirmed that rule. own measurementModbus RTU over RS-485, with and without demand on the emulated meter · 3000 W setting · 16 September 2026 manufacturerConow, e-mail of 28 September 2026 · 28 September 2026

It responds slowly. If your consumption changes faster than about half a minute, it makes the result at the meter worse than doing nothing: it trails the demand and delivers at the wrong moment. With consumption going up and down over a 20-second period, the error at the meter was 134%; at 10 seconds, 206%. Starting or reversing takes 12 to 35 seconds, adjusting while running 4 to 9, and through the meter it responds after about 4 seconds. own measurementour techbench, emulated meter with a varying demand, and step response over Modbus · 3000 W setting · 16 September 2026

Conow confirms the mechanism: the internal control loop is fast, but the power ramps up slowly on purpose, to spare the grid, and the meter adds a delay of its own. If your consumption suddenly drops, it can briefly feed back in those seconds, Conow writes. It is a design choice, not a defect. A faster start-up and an adjustable ramp are being looked at by Conow for later firmware, without a date. manufacturerConow, e-mail of 28 September 2026 · 28 September 2026

When your own control locks up. If your own control crashes, it keeps charging. We set it to charge at 500 W and then sent no further message, while the connection stayed open: for 178.6 seconds, 88 readings at the socket, and not one below 477 W. Conow had written beforehand that it goes to rest after 60 seconds without valid Modbus communication, also when the controlling process stops. own measurementour techbench, at the socket, watchdog test · 3000 W setting · 19 September 2026 manufacturerConow, answer to our questionnaire, repeated in the e-mail of 18 September 2026 · 18 September 2026

It later turned out to be two timers. The bridge between the network and the RS-485 port kept the connection open, and as long as it is open, the Atlas carries on with the last command. When we cut the connection ourselves, it did stop: charging from 1998 W was at zero within 60.2 seconds, discharging from 993 W within 61.9 seconds. After that it does not restart on its own: according to Conow you have to write the command again. Conow commits to a configurable timeout in future firmware, without a date. own measurementour techbench, at the socket, connection cut twice · 3000 W setting · 22 September 2026 manufacturerConow, e-mail of 28 September 2026 · 28 September 2026

For you that means: controlling it yourself works, locally and without a cloud, as long as your control keeps talking. If it goes quiet while the line stays open, it carries on with the last command. If the line drops, it stops and you start again.

What the documentation did not say. Control it yourself on this firmware and you run into errors you only see when something goes wrong. In the correspondence Conow acknowledged seven; that count is ours, Conow gives no total. manufacturerConow, e-mails of 16 to 28 September 2026 · 28 September 2026

• Leave the maximum power field empty in a discharge command and it does not deliver the default 800 W, but its full power. This is register 10101. Conow called it "a huge bug" and commits to fixing it as soon as possible, without a date. manufacturerConow, e-mail of 17 September 2026 · 17 September 2026
• The lower limit for the state of charge also accepted zero over Modbus, which bypassed the battery's protection (register 10102). Conow enforces a lower limit in the mid-October firmware. manufacturerConow, e-mails of 17 and 18 September 2026 · 18 September 2026
• The same error also dropped the lower limit for forced charging. manufacturerConow, e-mail of 18 September 2026 · 18 September 2026
• The app and the register disagreed on the reserve: the app showed one value, and the battery stopped at the other. manufacturerConow, e-mail of 18 September 2026 · 18 September 2026
• The rule by which it picks up a charge command was not in the documentation; Conow commits to describing it. manufacturerConow, e-mail of 18 September 2026 · 18 September 2026
• The datasheet gave different voltage ranges for the battery; Conow corrected that, see the contradictions below. manufacturerConow, e-mails of 16 and 17 September 2026 · 17 September 2026
• The "Modbus TCP" label in the app, see above. manufacturerConow, e-mail of 17 September 2026 · 17 September 2026

There is more. A read that cuts a two-register field in half now silently returns a wrong value; from mid-October it returns an error, according to Conow. Which firmware is installed cannot be read over Modbus; a register for that comes in the next update. According to Conow, mid-October also brings the correct nominal capacity in the firmware, the anti-feed-in settings over Modbus, and a register for instantaneous DC power, which Conow adds for measurements like these. manufacturerConow, e-mails of 17, 18, 22 and 28 September 2026 · 28 September 2026

Home Assistant. Not tested. Tuya Smart wrote that Modbus and Home Assistant follow the same path on the Atlas as on the Lyra from the same brand. Whether an existing integration recognises this unit, I do not know. manufacturerTuya Smart, e-mail of 8 September 2026 · 8 September 2026

The Tuya ecosystem. Not tested. Conow sells the Atlas as "Works with Tuya". According to the product page, you control it with Conow's own app or with the Tuya or Smart Life app, and it works with other devices in that ecosystem, "from smart plugs and lights to heat pumps and blinds", and with Shelly and everHome. Whether it responds to another device in your home through such a link, and how fast, we did not measure: this test ran through Conow's own app and over Modbus. It comes back in the revisit after the mid-October firmware. manufacturerConow, Atlas 6000 AC product page, read on 4 October 2026 · 4 October 2026

Connecting it. A residual-current device belongs with it. Conow recommends a 30 mA type A, and says the Atlas monitors DC leakage current itself, from 6 mA. We did not test the latter. manufacturerConow, e-mail of 28 September 2026 · 28 September 2026

The backup port. It works: I charged another battery from it, and that went fine. We did not measure its power or its switchover time. Conow gives 3000 W continuous, 4500 VA peak for 3 seconds and a switchover within 10 ms; this bench has no setup for island operation. own measurementobserved in use, not measured · not recorded · 3 October 2026 manufacturerAtlas 6000 AC datasheet, backup port row · 10 September 2026

Heat, noise and build

Where you put it. Place it free-standing, not in a cupboard, and keep fabric away from the cooling fins. At 3000 W continuous, over a full discharge, the cooling fins reached 60 to 62 °C (thermal camera) and the battery reported 43.8 °C at the end (register). Fins and battery differ a lot, and the register says nothing about the radiator. own measurementthermal camera at about half a metre, and the device's temperature register; full discharge · 3000 W setting · 18 September 2026

That the derating responds to the battery temperature and not to the fins, we infer: during that discharge the fins were well above 50 °C without the power dropping, and the warm charge dropped exactly when the battery reached 45.7 °C. We have asked Conow which sensor decides. derivedfrom the fin temperature from the thermal camera, the power during the full 3000 W discharge of 18 September, and the warm charge of 20 September

At full power you can hear it. At 3000 W continuous it makes a high-pitched whine, not on the lower settings. There was no sound level meter, so there is only the recording. Other batteries on the same bench, in the same room, made no noticeable tone at full power. own measurementheard during the measurements of 14 to 21 September, recorded with a phone, not measured · 3000 W setting · 21 September 2026

The specifications, with origin

The rows a buyer compares. eigen meting (own measurement) carries the setting in the last column; fabrikant (manufacturer) the document.

Specification	Value	Origin	Condition or document
Nominal energy	6028 Wh	manufacturer	datasheet, received 2026-09-08, archived 2026-09-10
Usable according to Conow	6000 Wh	manufacturer	Conow e-mail, 2026-09-08
Delivered at the socket, 100 to 10%	5192.6 Wh at 1500 W; 3.4% less at 3000 W	own measurement	inline kWh meter, 3000 W setting, 2026-09-17 and 2026-09-18
Charging power	3061 W	own measurement	at the socket, 3000 W setting, 2026-09-16
Discharging power	3006 W	own measurement	at the socket, 3000 W setting, 2026-09-16
Stated power	3000 W charging and discharging; 800 W on the grid by default	manufacturer	datasheet
Power tolerance	1% of the power plus 1% of 3000 W: 2940 to 3060 W	manufacturer	Conow e-mail, 2026-09-28
Round trip at 1500 W	0.87; charging 91.3%, discharging 95.0%	own measurement	socket to socket, 3000 W setting, 2026-09-18
Round trip at 3000 W	0.86 after a cool start, 0.81 after a warm one	own measurement	socket to socket, 3000 W setting, 2026-09-20
Efficiency per step, 100 to 3000 W	0.754 (charge) and 0.788 (discharge) at 100 W; 0.893 to 0.969 between 300 and 1000 W, error bars overlap; 0.980 to 0.921 from 1500 to 3000 W; 200 W charge without a figure	own measurement	the device's energy counter, ladder-v1 and ladder-low-v2, 3000 W setting, 2026-09-21
Promised round trip	at least 81%	manufacturer	Conow e-mail, 2026-09-08
Derating when charging	above 45 °C to about 2590 W, above 50 °C to about 1940 W	manufacturer	Conow e-mails, 2026-09-08 and 2026-09-17
Warm charge, measured	3039 W up to 45.7 °C, then 2623 W	own measurement	inline kWh meter, 3000 W setting, 2026-09-20
At 30 W requested	27.32 W at the socket	own measurement	inline kWh meter, 3000 W setting, 2026-09-20
At rest, grid side	0.0 W over eight hours	own measurement	inline kWh meter, idle, 2026-09-19
At rest, from the battery	under 5 W; no more precise figure	own measurement	inline kWh meter, idle, 2026-09-19
At rest according to Conow	less than 10 W with Wi-Fi on	manufacturer	Conow e-mail, 2026-09-08; not on the datasheet
Cooling fins at 3000 W continuous	60 to 62 °C	own measurement	thermal camera, 3000 W setting, 2026-09-18
Battery temperature at the end of that discharge	43.8 °C	own measurement	device register, 3000 W setting, 2026-09-18
Battery voltage	16.2 to 21.6 V according to Conow; measured 18.55 V at 8% and 20.47 V at 80%	manufacturer · own measurement	Conow correction; device registers, 3000 W setting, 2026-09-17
Backup port	3000 W continuous, 4500 VA peak for 3 s; switchover within 10 ms	manufacturer	datasheet; the port works in use, power and switchover time not measured
Residual-current device	30 mA type A recommended; own DC leakage monitoring from 6 mA	manufacturer	Conow e-mail, 2026-09-28; not tested
Chemistry	LFP	manufacturer	datasheet
Communication	Wi-Fi, Bluetooth LE, RS-485 (Modbus RTU), LAN	manufacturer	datasheet; Modbus over RS-485 used by us, Modbus TCP not present
Cloud needed for control	no, over RS-485	own measurement	Modbus RTU through a network bridge, 2026-09-16
Firmware	Main V1.0.0 (1.0.8) / MCU V1.01.06; no firmware register	own measurement	read from the app, 2026-09-21
Ingress protection	IP65	manufacturer	datasheet
Dimensions and weight	800 x 315 x 236 mm, 55.8 kg net	manufacturer	datasheet
Cycles	10,000 to 70% at 25 °C · more than 6000 at 25 °C	manufacturer · third party	Conow datasheet · the sheet at Elektramat; see the contradictions below
Warranty	10 years	manufacturer	datasheet
Price	€ 1,720.96; € 285 per kWh on the datasheet, € 331 per kWh on what it delivered	third party · derived	Elektramat.nl, including VAT, in stock, checked 2026-10-03; per kWh on 6028 Wh and on 5.19 kWh

Where the sources contradict each other

There is a Conow datasheet without version or date, a manual, Conow's e-mails, and a second sheet that Elektramat offers with the Atlas. They disagree on five points. I do not choose; I write down what each document says, and where our measurement adds something, the measurement stands next to it.

Usable capacity. The datasheet gives a depth of discharge of 100% on 6028 Wh. Conow wrote 6000 Wh usable in the e-mail of 8 September, with an adjustable reserve of 10 to 100%, and on 17 September that the firmware gets the correct nominal capacity in mid-October. Our measurement does not settle this, because it measures something else: 5192.6 Wh at the socket, from 100 to 10%, at 1500 W. manufacturerAtlas 6000 AC datasheet; Conow, e-mails of 8 and 17 September 2026 · 17 September 2026 own measurementour techbench, inline kWh meter · 3000 W setting · 17 September 2026

Battery voltage. The datasheet of 10 September says 13.0 to 21.9 V. In its e-mails Conow corrected that to 16.2 to 21.6 V, after earlier sheets with yet other ranges. Here the measurement settles it: we read 18.55 V at 8% state of charge and 20.47 V at 80%, both within the corrected band. manufacturerAtlas 6000 AC datasheet; correction in Conow's e-mails · 17 September 2026 own measurementdevice registers, start and end of a charge · 3000 W setting · 17 September 2026

Cycles. Conow's datasheet says 10 years or 10,000 cycles, at 25 °C to 70% remaining capacity. The sheet Elektramat offers with the Atlas, not a Conow document, says more than 6000 cycles at 25 °C, without a remaining capacity. They are not even the same kind of claim: without a floor, a cycle count says little. We have no third figure to add; you do not measure cycles in eight days. manufacturerAtlas 6000 AC datasheet, general parameters row · 10 September 2026 third partyAtlas 6000 AC datasheet on elektramat.nl · 15 September 2026

The backup port. On the datasheet every row for the backup port is about output: 3000 W continuous, 4500 VA peak, 13.7 A. The wiring diagram in the manual also shows an existing balcony solar system on that same port, and a balcony solar system feeds power in. How much the port accepts is in neither. manufacturerAtlas 6000 AC datasheet and Dutch manual v3, wiring diagram · 10 September 2026

Control. The app calls the switch "Modbus TCP", and it is RTU over RS-485. And Conow wrote beforehand that it goes to rest after 60 seconds without valid communication, also when the controlling process stops; on our bench it kept charging as long as the connection stayed open. That is covered above under control, with the explanation Conow gave later. own measurementour techbench, at the socket, watchdog test · 3000 W setting · 19 September 2026

What I don't know yet

What is here belongs to the test. Each point with the reason, when it can be done, and what it would decide. They only go when a measurement closes them.

• Standby consumption as a figure. In eight hours idle the counter moved two steps, and the error on that includes zero. When: after about 80 hours idle, or with the DC register. Decides: what standing still costs per year.
• The switchover time to backup power. The port works, but there is no setup for island operation. When: undetermined. Decides: whether Conow's "within 10 ms" holds.
• Everything the mid-October firmware changes. Not out yet. When: mid-October 2026, in the revisit. Decides: whether it keeps charging when your control goes quiet, and whether the acknowledged errors are fixed.
• Warm charging on the 800 W setting. The derating was measured only at 3000 W, and 800 W is the out-of-the-box setting. When: in the revisit. Decides: whether warm charging also costs for anyone who does not unlock the setting.
• A warm location. Our warm charge got its heat from a discharge, in a cool room. When: undetermined. Decides: whether an attic or a cupboard gives the same loss.
• Which sensor drives the derating. Conow wrote that the system monitors several temperature sensors on the battery pack, but there is no register that reports the derating. When: with Conow's answer. Decides: whether you can read the derating in advance.
• What it does when its meter drops out. According to Conow it falls back to the configured base load after a minute without meter data, and to zero without a base load. We did not measure that. When: undetermined. Decides: whether it stops safely when the meter is lost.
• Cycles. 10,000 and more than 6000 are two claims without the same floor, and this bench does not measure lifetime. When: not on this bench. Decides: whether Conow's 10,000 cycles hold.
• Who repairs it in the Netherlands. That is not in what Conow sent us. When: after asking in writing. Decides: where you can go when it breaks.
• Whether the protocol documentation has been updated. Conow wrote several times that it puts the corrections on GitHub. Whether that happened, we did not check. When: before the revisit. Decides: whether you can build your own control on the documentation.
• Control through the Tuya ecosystem. Conow advertises that the Atlas works with other Tuya devices, and with Shelly and everHome. We controlled it only through Conow's own app and over Modbus. When: in the revisit after the mid-October firmware. Decides: whether it can respond to other devices in your home without a script of your own.

One point that was open on the card is now a measurement: the round trip at 3000 W, measured on 20 September with a cool and a warm charge, and written out above. own measurementour techbench, inline kWh meter · 3000 W setting · 20 September 2026

Closing

What follows can be traced back to a measurement above. I do not give a judgement; the judgement is yours.

Pro. It delivers slightly more power than promised, 3061 W charging and 3006 W discharging, and follows a command closely, never above the request when discharging. The round trip is within Conow's promise: at 1500 W, 87 of every 100 kWh come back. At 30 W requested, 27.32 W still arrives at the socket, at rest you do not see it on your meter, and you can control it yourself locally, over RS-485 and without a cloud. On top of that, a manufacturer that answered in three rounds, acknowledged seven errors and committed to fixes. own measurementour techbench, the findings above · 3000 W setting · 21 September 2026

Con. Warm charging derates: halfway through, the charging power drops to 2623 W, and the charge costs 5.9% more power. Efficiency drops at low power: at 100 W, 0.754 charging and 0.788 discharging, and Conow itself states 30 to 50 W of fixed loss. Out of the box it is set to 800 W. With Modbus the whole control loop is yours, and if your control goes quiet while the line stays open, it keeps going. It does not follow consumption that changes faster than half a minute. And at full power you hear a tone. own measurementour techbench, all findings above · 3000 W setting · 21 September 2026

The price against the market. € 1,720.96 is € 285 per kWh on the datasheet's 6028 Wh, and € 331 per kWh on the 5.19 kWh it delivered at the socket on our bench. On 3 October 2026 the band for 5 to 6 kWh at Dutch shops, in stock, ran from € 1,218.95 to € 1,799. The Atlas sits at the top end of that band. The neighbours and the competitor are below. derivedfrom the price of 2026-10-03, the datasheet's 6028 Wh and the measured 5192.6 Wh third partyshop pages, see the table under alternatives · 3 October 2026

Who it is for: out of the box it is set to 800 W, and on the lower settings the tone was not audible in any measurement. Set it to 3000 W in the living room and you hear it at full power. Controlling it yourself works locally, if your own control watches the connection; if you do not want that, leave the switch off and the app handles it. Charge often at 3000 W right after a heavy discharge and you get less back than with a cool charge. If your house often asks for little, the fixed loss Conow states, 30 to 50 W, weighs most; at rest you do not see it on your meter. And if your consumption changes faster than half a minute, its control on the meter does little for you in those seconds. own measurementour techbench, the findings above · 3000 W setting · 21 September 2026 manufacturerConow, e-mail of 28 September 2026 · 28 September 2026

Alternatives

We checked the prices in the band on 3 October 2026, on the shop page itself, including VAT. A price that was not on the page is not here; a price is a reading of one day. third partythe shop pages in the table, including VAT · 3 October 2026

Device	Shop	Price	Stock	Origin
Conow Atlas 6000 AC, 6028 Wh	Elektramat.nl	€ 1,720.96	in stock	third party, 2026-10-03
Growatt Aura 5000, 5 kWh	Solar Outlet, with P1 meter	€ 1,218.95	in stock	third party, 2026-10-03
Growatt Aura 5000, 5 kWh	plugin-batterij.nl	€ 1,225	in stock	third party, 2026-10-03
EcoFlow STREAM AC 5000, 5 kWh	ecoflow.nl	€ 1,549	in stock	third party, 2026-10-03
EcoFlow STREAM 5000, 5 kWh	ecoflow.nl	€ 1,799	pre-order, delivery end of October	third party, 2026-10-03
AEG SolarCube Pro (AS-BBL08-6K), 6.0 kWh	aegthuisbatterij.nl	€ 1,799	in stock	third party, 2026-10-03
Samduo NEX-E6000, 6 kWh	stekker-batterij.nl	€ 1,799	available, delivered 6 October	third party, 2026-10-03
Anker SOLIX Solarbank 4 E5000 Pro, 5 kWh	thuisaccu.nl	€ 1,799	in stock	third party, 2026-10-03
AEG AS-BBL09, 4.8 kWh	shop.thuisbatterijnederland.nl	€ 1,520	in stock	third party, 2026-10-03
Voltdeer SR5000, 5.12 kWh	multi-solar.nl	€ 1,199	sold out	third party, 2026-10-03
Bluetti Balco 500, 5 kWh	no shop found	no price	—	third party, stekkerdeal.nl listed no shop, 2026-10-03

Cheaper in the band, and in stock, is the Growatt Aura 5000, from € 1,218.95. The EcoFlow STREAM AC 5000 costs € 1,549. At € 1,799 four sit side by side: the AEG SolarCube Pro, the Samduo NEX-E6000, the Anker SOLIX Solarbank 4 E5000 Pro, and the EcoFlow STREAM 5000 on pre-order. The Voltdeer SR5000 at € 1,199 was sold out, and for the Bluetti Balco 500 we found no shop price. The AEG AS-BBL09 at € 1,520 has 4.8 kWh and falls just below the band. For the Atlas, p1meter.nl reports an introductory price of € 1,511 in October, but on 3 October no shop showed that price. third partythe shop pages in the table; p1meter.nl for the announced introductory price · 3 October 2026

Where it sits in the price band5 to 6 kWh, shop prices of 3 October 2026

The competitor in the band is the AEG SolarCube AS-BBL08-3K, sold as the SolarCube Pro, for € 1,799, measured on the same bench. Over a closed cycle on the 800 W setting, 81.29% came back on the AEG. That figure does not stand on the same footing as the 0.87 above: the AEG was measured with an earlier version of our method, on a different setting and at a different power, which is also why it is not in the round-trip chart. Where they do differ directly: at 30 W requested, the Atlas still delivers 27.32 W, while the AEG is a net importer below 41 W. Both speak Modbus over RS-485. own measurementour techbench, AEG SolarCube Pro, closed cycle · 800 W setting · 12 September 2026 third partyaegthuisbatterij.nl, in stock · 3 October 2026

Frequently asked questions

Related articles

• The review of the AEG SolarCube Pro, the competitor in the band, on the same bench
• Controlling a home battery per brand for what local, cloud and Modbus mean per brand
• The error in home battery cycle calculations for why a cycle count without a floor is not a figure
• Home battery payback time tested (in Dutch) for what you can and cannot calculate with a measured round-trip efficiency

Transparency

Conow sent me this Atlas 6000 AC; the contact ran through Tuya Smart. The conversation started about their smaller Lyra 2500, and Conow proposed the Atlas itself. No money was paid. They did not see the script, nor the video before publication, nor this article. There are no agreements about what I say about it. I received the device free of charge and may keep it.

During testing we had technical contact: I sent what the bench measured, Conow answered, and according to Conow part of that goes into the mid-October firmware. What Conow wrote appears in this article where it touches a finding, as Conow's statement and with the date of the e-mail.

This article contains no affiliate links and no discount code. The prices are shops' asking prices on the stated dates, not recommendations, and I earn nothing if you buy anywhere.

Every figure in this piece carries its origin in the line: eigen meting (own measurement) with the grid setting, the instrument and the day; fabrikant (manufacturer) with the document; derde (third party) with the source; afgeleid (derived) with what it follows from. The charts draw the same measurements, with the origin in the caption. The video and this article come from the same review card, and a gate in the pipeline refuses a figure that is not on that card. What I could not measure is under "What I don't know yet", not tucked away in a subordinate clause.

Sources