18650 vs 21700 Battery Cells: Which Is Better for a Custom Battery Pack?

Moving a design from 18650 to 21700 cells looks like a free upgrade on paper. Same chemistry family, same nominal voltage, more energy per cell, fewer parts to weld. Then the enclosure comes back 5 mm short, the parallel count drops from ten strings to six, and the pack that used to shrug off a stalled-motor event starts tripping on overtemperature.

The 21700 vs 18650 question is not a capacity comparison. It is a pack architecture decision that touches your enclosure, your interconnect cross-section, your thermal path, your redundancy, and your cost per watt-hour. The cell you pick sets constraints you will live with for the life of the product.

Last reviewed: August 2026. Cell model references reflect products available at time of writing. Always confirm against the current manufacturer datasheet.

What the 18650 and 21700 names actually tell you

Both names are dimensional codes. An 18650 cell has an 18 mm diameter and a 65 mm length. A 21700 has a 21 mm diameter and a 70 mm length. The trailing digit is a legacy marker for cylindrical geometry and carries no information you need.

What the name does not tell you is everything that matters: chemistry, capacity, internal resistance, continuous current rating, cycle life, or temperature limits. Two cells with the same four-digit code can differ by a factor of three or more in continuous discharge capability. Treating "18650" as a spec is the single fastest way to end up with a pack that meets its energy target and fails its power target.

One practical note that catches people in mechanical design. Datasheets publish a maximum envelope, not a nominal size. Samsung's INR21700-50S specification gives a maximum cell envelope of roughly 21.25 mm by 70.62 mm rather than a clean 21 by 70. Add the wrap, the holder wall, and clearance for venting, and your array pitch grows further. Dimension the enclosure from the datasheet maximum plus stack-up tolerance, never from the name. And set aside protected cells with an integrated PCB on the negative end: those are consumer replacement parts, several millimetres longer than the bare cell. An OEM pack uses unprotected flat-top cells with protection handled at pack level by the [Internal Link: Battery Management System Design for OEM Packs].

18650 vs 21700 dimensions and what they do to your enclosure

Run the volumes. An 18650 occupies roughly 16.5 cm³. A 21700 occupies roughly 24.2 cm³. One 21700 takes up about 1.5 times the space of one 18650 and stores roughly 1.6 to 1.7 times the energy in current commercial cells. That gap is the whole argument for the larger format, and it is smaller than most people assume.

Here is an illustrative comparison for a 36 V nominal pack targeting about 1 kWh. Using a 3.0 Ah 18650, you land on 10S10P: 100 cells, around 1,080 Wh. Using a 4.9 Ah 21700, you land on 10S6P: 60 cells, around 1,060 Wh. Same voltage, same energy, 40 cells fewer.

Values are representative ranges for common commercial cells. Verify all figures against the specific cell datasheet before committing to a design.

With typical holder pitch, that 60-cell 21700 array occupies somewhere in the region of 10 to 15 percent less total volume than the 100-cell 18650 array, and comes in roughly 10 percent lighter. Meaningful, not transformative. And it arrives with a hard geometric constraint: the pack is at least 5 mm taller in the cell axis, and each column is 3 mm wider.

That height difference decides more programs than energy density does. If the pack has to drop into an existing enclosure, an existing tool platform, or an existing under-deck cavity, the conversation is over before it starts. Retrofit programs stay on 18650 regardless of what the density numbers say.

Energy density, capacity, and the "21700 5000 mAh" question

The 21700 wins on both gravimetric and volumetric energy density, but by less than the size difference suggests. At the same chemistry generation, expect single-digit to low-double-digit percentage gains in Wh/kg and Wh/L. The mechanism is simple: the steel can, header, vent, and current interrupt device are fixed overhead, and that overhead is a smaller fraction of a bigger cell.

Now the part that costs people runtime in the field. A cell marketed as a 5000 mAh 21700 rechargeable battery frequently carries a rated minimum below 5000 mAh. Samsung's INR21700-50E datasheet specifies a minimum of 4900 mAh. The INR21700-50S is commonly sold as 5000 mAh typical with a rated minimum around 4800 mAh. Independent bench testing of the 50E has come in slightly under 5000 mAh at low discharge rates, which is exactly what the datasheet says should happen.

The engineering habit that protects you: size the pack on rated minimum capacity, at your actual discharge rate, at your coldest specified operating temperature, at your end-of-life threshold. If your marketing sheet promises eight hours of runtime and you sized on typical capacity at 23 °C at 0.2C, you will get warranty claims in the second winter. That is not a cell problem. That is a specification problem.

The same discipline applies to 18650 battery specs, where typical and minimum values also diverge, and the gap widens at higher C-rates.

Discharge capability: the bigger cell is not automatically the higher-current cell

This is where the intuition breaks, and it is the most useful thing in this article.

Compare two real cells. The Molicel INR-18650-P28A is a 2.8 Ah 18650 rated at 35 A continuous. The Samsung INR21700-50E is a 4.9 Ah 21700 rated at 9.8 A continuous, which is 2C. The smaller cell delivers roughly 3.5 times the current of the larger one. Cell design intent beats cell size, every time.

Even comparing a high-power 21700 against a high-power 18650, the per-cell current advantage is modest relative to the volume difference. And headline current numbers often carry conditions buried in the datasheet. The Samsung 50S is rated 25 A continuous without a temperature cutoff, and up to 45 A only when managed with an 80 °C cell temperature cutoff. If your pack cannot hold the can below 80 °C at that current, the 45 A figure is not available to you. Read the footnotes on every current rating you plan to design around.

Now scale it to the pack, which is the number that actually matters. Take the same two architectures at roughly equal energy:

• 10S10P of a 35 A high-power 18650: approximately 1,010 Wh, with a theoretical 350 A continuous ceiling.
• 10S6P of a 25 A high-capacity 21700: approximately 1,080 Wh, with a theoretical 150 A continuous ceiling.

Comparable energy, more than double the current headroom on the 18650 side. Because the smaller format needs more parallel strings to reach the same energy, and current capability scales with parallel count, 18650 packs frequently come out ahead on power density even though the 21700 wins on energy density.

Both of those ceilings are datasheet maxima measured on single cells in controlled conditions. Inside a holder, surrounded by neighbours, with limited airflow, derate hard. A working rule of thumb is to design continuous operation well below the datasheet maximum and reserve the top of the range for short peaks with a defined duty cycle.

The decision rule I use: if the application is power-limited, meaning high peak current relative to stored energy, 18650 high-power cells with a high parallel count usually give the cleaner design. If it is energy-limited, meaning long runtime at modest C-rate, the 21700 is the better answer.

Thermal management and why 21700 packs run hotter in the core

Physics does not care about your BOM cost. For a cylinder with roughly uniform internal heat generation, the temperature rise from core to surface scales with the square of the radius. The 21700 radius is about 17 percent larger, which puts the core-to-surface gradient roughly 35 percent higher at the same volumetric heat generation rate. The 21700 also has less lateral surface area per unit of internal volume: less area to reject heat through, further for that heat to travel. At low C-rates, irrelevant. At 2C continuous in a sealed enclosure, it drives your design.

Three consequences that show up in real packs:

Your temperature sensors under-report by more. An NTC bonded to a cell can is reading the coolest part of the cell, and in a 21700 array the gap between what the sensor sees and what the jelly roll core is doing is larger. Place sensors on the worst-case cells, meaning the interior of the array and downstream of any airflow, not the accessible edge cells.

Cold charging protection matters more. Charging below roughly 0 °C drives lithium plating and permanent capacity loss. A larger-diameter cell takes longer to warm through to the core, so a cold 21700 pack needs a more conservative warm-up interlock before charge current is allowed.

Gradient control gets harder. A good target is holding the spread across cells under about 5 °C. As a rough guide, the Arrhenius relationship puts aging rate at roughly double for every 10 °C increase, so uneven packs age unevenly and the weakest group sets end of life.

None of this makes the 21700 a worse cell. It means the thermal design work is not optional at the same C-rates you got away with on 18650. More on that in [Internal Link: Battery Pack Thermal Management for Industrial Equipment].

Impact on pack design: series-parallel, BMS, welding, and enclosure

Series count is set by system voltage and is identical for both formats. A 36 V nominal pack is 10S whether you build it from 18650 or 21700 cells. That has a cost implication people miss: your [Internal Link: Battery Management System Design for OEM Packs] channel count, balancing topology, and sense harness do not get cheaper when you move to 21700. The BMS scales with series count, not cell count.

What does change:

Weld count. Sixty cells instead of a hundred means roughly 40 percent fewer weld joints. Every weld is a candidate for an open circuit or a high-resistance joint that surfaces as a thermal hotspot two years later. On low and mid-volume programs that is real assembly time and real yield risk removed.

Redundancy. In a 10P group, losing one cell removes 10 percent of the current path. In a 6P group it removes 17 percent, and the survivors carry proportionally more current, run hotter, and age faster. That matters on long-service-life products with no field access to the pack. Cell-level fusing is also easier to justify at higher parallel counts.

Interconnect cross-section. Each cell in a 6P group carries more current than each cell in a 10P group at the same pack load. Do not reuse the nickel strip or busbar cross-section from the 18650 design. Recalculate for the new per-cell current and check joint temperature rise under peak load.

Capacity granularity. Stepping 10P to 11P on 18650 adds about 10 percent energy. Stepping 6P to 7P on 21700 adds about 17 percent. If your runtime spec sits between two 21700 steps, you either miss the target or pay for surplus energy. Finer resolution is an underrated argument for the smaller format on tightly specified products.

Which OEM projects should stay on 18650

Retrofit and replacement programs, first and always. If there is an existing enclosure, an existing battery bay, or an existing tool platform, the geometry decides.

Beyond that:

• Shallow or thin enclosures where 5 mm of cell height cannot be found. Handheld instruments, slim housings, under-panel mounting.
• High peak current relative to energy. Power tools, robotics with stall events, drones, motorised actuators with high inrush.
• Products with tight runtime specifications that benefit from finer parallel-count resolution.
• Long-service-life equipment where redundancy across many parallel cells is a stated design requirement.
• Regulated or long-qualification industries where an already-qualified cell is in the design and requalification cost far exceeds the density gain.
• Genuinely low-volume programs, where the wider distributor availability of an 18650 lithium battery pack build across many cell brands is easier to secure in small quantities.

Anyone telling you the 18650 is obsolete has not looked at what cell manufacturers are shipping. Molicel introduced the INR-18650-P30B as a new-generation high-power 18650 with a significantly reduced DC resistance compared with the P28A and a full-range fast charge rate raised from 2C to 3C. That is fresh engineering investment in the format, not legacy inventory.

Which OEM projects should move to 21700

Greenfield designs where the enclosure is still on the screen and nobody has committed to a cavity dimension.

• Energy-limited applications where runtime is the headline spec and continuous C-rate is modest.
• Weight or volume constrained products per watt-hour delivered. Light EV, mobility, portable field equipment.
• Higher-energy packs where cell count and weld count are meaningful cost drivers.
• Programs with a long production life. Cell development investment is concentrated on 21700 and larger formats, so a future capacity upgrade inside the same mechanical envelope is more likely to exist. Swapping a 4.9 Ah cell for a 5.2 Ah cell in the same footprint is a minor change. Changing formats is a new pack.

Format-specific build details sit on the [Internal Link: Custom 18650 Battery Pack] and [Internal Link: Custom 21700 Battery Pack] pages.

What about 4680 and the 46-series cells?

The 46-series format, 46 mm diameter with lengths around 80 mm, carries roughly five times the volume of a 21700 and was developed for EV pack integration with tabless electrode construction. For automotive-scale programs it is a serious direction.

For custom pack work below a few kWh, it is not yet the practical answer. Sourcing in modest quantities remains difficult outside automotive supply agreements, and the granularity problem gets much worse: one cell is a very large energy step, and low cell counts make common industrial system voltages awkward to reach. For most industrial, mobility, and portable OEM programs, the real choice today is still 18650 or 21700.

Cell selection checklist for OEM battery pack projects

Run your project against this before you specify a format. If you cannot answer an item, that gap is your next task, not something a supplier can fill in for you.

1. Nominal and maximum system voltage, which fixes your series count.
2. Usable energy required at end of life, at your coldest specified operating temperature, at your actual discharge rate.
3. Continuous current, peak current, peak duration, and duty cycle.
4. Available envelope in all three axes, including holders, interconnect, BMS board, wiring, and clearance.
5. Weight budget for the assembled pack.
6. Operating temperature range, allowed charge temperature range, and the thermal path available (conduction to chassis, forced air, or nothing at all).
7. Cycle life target and the end-of-life criterion you will be measured against, commonly 80 percent of initial capacity.
8. Required charge time, which sets your charge C-rate and your cold-charge policy.
9. Certification path, including standards such as IEC 62133 and UN 38.3, plus any application-specific requirements. See [Internal Link: Battery Testing and Certification].
10. Annual volume and forecast, which determines whether a dedicated cell allocation is realistic.
11. Service model: replaceable pack, field-serviceable, or sealed for life.
12. Retrofit constraint: does this pack have to fit something that already exists?

Item 12 overrides items 1 through 11 more often than anyone likes to admit.

At PackForge Energy we work through this list with engineering teams before quoting, because a format decision made on energy density alone tends to get reversed once the mechanical review happens. If you are developing a battery-powered product, the information that gets you a useful engineering response is: nominal voltage, required capacity, continuous and peak current with duty cycle, available dimensions, connector and communication requirements, operating temperature range, and expected annual quantity. Send that and you get an evaluation. Send "we need a 36V pack" and you get a placeholder price with risk padding in it, which helps nobody. [Internal Link: Request an Engineering Evaluation]

Frequently asked questions

Is a 21700 always better than an 18650?

No. The 21700 offers higher energy density and reduces cell and weld count for the same pack energy, which makes it the stronger default for new, energy-limited designs. The 18650 remains better for retrofits into existing enclosures, for shallow housings where 5 mm of extra height is unavailable, for high-peak-current applications where a high parallel count of power cells is needed, and for programs where an existing cell qualification is expensive to repeat.

Can I replace 18650 cells with 21700 cells in an existing battery pack?

Rarely without redesign. A 21700 is 3 mm wider and 5 mm longer, so array pitch, holder geometry, and overall enclosure height all change. Beyond the mechanics, the parallel count usually drops, which changes per-cell current, interconnect cross-section requirements, thermal behaviour, and pack current capability. Series count and therefore BMS configuration typically stay the same, but the mechanical and interconnect design should be treated as new work.

What is the real capacity of a 21700 li-ion 5000 mAh cell?

Usually slightly less than the marketing figure. Cells sold as 5000 mAh often specify a rated minimum below that, and manufacturer datasheets are the authoritative source. Samsung's INR21700-50E specifies a 4900 mAh minimum, and the INR21700-50S is typically quoted at 5000 mAh typical with a lower rated minimum. Size your pack on the minimum rated capacity at your actual discharge rate and temperature, then apply your end-of-life derating.

Which recommended 18650 battery should I use for an OEM pack?

There is no single answer, because 18650 cells split into energy-optimised and power-optimised families. Energy cells in the 3.0 to 3.6 Ah class suit long-runtime applications at modest C-rate. Power cells in the 2.5 to 3.0 Ah class with continuous ratings of 30 A and above suit high-current and high-peak applications. Decide which constraint governs your product first, then shortlist cells from that family and validate against the datasheet.

How many cells do I need for a custom battery pack?

Series count comes from voltage: divide nominal system voltage by cell nominal voltage, typically 3.6 or 3.7 V for lithium-ion. Parallel count comes from energy: divide required usable energy by the energy of one series string, using minimum rated capacity rather than typical, then verify that the resulting parallel count also supports your continuous and peak current. Current capability, not energy, is the binding constraint more often than expected.

Do 18650 and 21700 packs need different BMS designs?

The BMS scales with series count, not cell count, so a 10S pack needs the same channel count in either format. What changes is the current path. Lower parallel counts in a 21700 pack mean higher per-cell current, so interconnect cross-section, current sense range, and fault thresholds should be recalculated. Temperature sensor placement also needs revisiting, since larger-diameter cells show a bigger difference between surface and core temperature.