Tesla Regen Energy Capture vs State of Charge: NCM/NCA Halves at Full Charge, LFP Does Not

BatterMachine, August 2026.

Every Tesla driver has seen regenerative braking reduced after a full charge. The dashed line appears on the power meter, the car coasts where it would normally slow, and the friction brakes do the work for the first part of the drive.

The mechanism is well understood. What is less clear is the magnitude: how much energy is actually lost, at what state of charge the loss becomes material, and whether the two cell chemistries in the Tesla fleet behave the same way. We measured all three across Model 3 and Model Y, in both nickel manganese cobalt / nickel cobalt aluminium (NCM/NCA) and lithium iron phosphate (LFP) packs.

The short answer is that a full charge halves regen capture on NCM/NCA packs and does nothing measurable to LFP packs.

Why a nearly full pack refuses regen

Regenerative braking is a charging current. It pushes energy into the pack in the same direction a charger does, briefly and at high power.

How much current a cell will accept is governed by voltage, not by any abstract notion of fullness. Every lithium-ion cell has an upper voltage limit it must not exceed. When current flows into the cell, the voltage measured at its terminals rises above its resting value, the open-circuit voltage (OCV), by an amount that grows with the current and with the cell's internal resistance. Roughly:

terminal voltage under charge = OCV + current x internal resistance

The battery management system (BMS) must keep that terminal voltage below the upper limit. Rearranging, the largest current it can allow is set by the gap between the upper limit and the present OCV, divided by the internal resistance. That gap is the headroom.

At a low state of charge (SOC), OCV sits far below the upper limit, the headroom is wide, and the pack will accept a hard regenerative braking event without approaching the limit. As the pack fills, OCV rises, the headroom narrows, and the allowable current falls with it. Near a full charge the headroom is nearly gone and the allowable regenerative current approaches zero. That is the dashed line on the power meter, and it retreats as you drive because OCV falls and the headroom returns.

The consequence of getting this wrong is why the BMS is conservative about it. Forcing terminal voltage past the upper limit causes lithium to deposit as metal on the anode instead of intercalating into it. That loss is irreversible, it degrades capacity permanently, and in the extreme it is a safety concern. So the limit is applied with margin rather than at the theoretical boundary.

Two of the three terms in that relation are roughly stable over a single drive at moderate ambient temperature. Internal resistance changes with temperature and current, but not dramatically within one journey. OCV is the term that moves, because it climbs steadily as the pack charges and falls as it discharges. So the behaviour of regen against SOC is essentially the behaviour of the OCV curve near the top of charge.

What the fleet shows

On NCM/NCA packs, regen capture is flat at about 24% of the energy the pack discharges, from mid SOC up to roughly 80%. Above that it falls steadily: 22.6% at 85% SOC, 21.3% at 90%, 17.4% at 95%, and 12.4% at 99%, with a 95% confidence interval of 11.4 to 13.4. A car leaving on a full charge recovers roughly half as much braking energy as the same car at a normal state of charge.

On LFP packs, capture is flat across the entire range. It measures 24.2% at 75% SOC and 22.8% at 99%, with a 95% confidence interval of 22.0 to 23.7. Whatever decline exists is confined to the last two points of charge and is small enough that it would not be noticeable from the driver's seat.

Why the chemistries differ

The difference follows from the shape of the two OCV curves.

An LFP cell holds a nearly constant voltage across most of its usable range. The curve is famously flat from roughly 30% SOC upward and only turns sharply upward in the last moments before full. So even at a displayed 100%, an LFP cell still sits some distance below its upper limit, and headroom remains.

An NCM/NCA cell behaves differently. Its OCV climbs steadily throughout the charge, so by the time the pack reads 90% the cell is already approaching its ceiling. The headroom has been shrinking for most of the charge rather than only at the end, which is why the capture curve begins to bend from about 80% rather than falling off a cliff at 99%.

This also explains why Tesla advises LFP owners to charge to 100% regularly while recommending that owners of NCM/NCA cars keep daily charging closer to 80%. The guidance matches what the packs physically do.

The state of power argument runs the other way

There is a complication worth confronting, because on cell capability alone the result should be the opposite of what we observe.

State of power (SOP) is the maximum power a pack can accept or deliver at a given moment. NCM/NCA cells have higher power capability than LFP cells. That is one of the reasons they are used in longer-range, faster variants. On charge SOP alone, an NCM/NCA pack should absorb regenerative braking at least as well as an LFP pack at every state of charge, not worse.

The mid-range behaviour is what makes the result interpretable. Through the middle of the SOC range the two chemistries are indistinguishable: 24.4% capture for both at 65% SOC, and 24.1% against 24.2% at 75%. If charge SOP were the binding constraint in ordinary driving, the chemistry with more power capability would capture more energy. It does not. That tells us that in normal driving the limit on how much energy comes back is not the pack at all. It is how much decelerative energy the driver generates in the first place.

The two chemistries separate only where charge SOP does become binding, and it becomes binding far earlier on NCM/NCA. So the difference between them is not about power capability. It is about how much voltage headroom remains at a displayed full charge.

Two explanations fit that observation and fleet telemetry cannot separate them. Tesla may reserve more capacity margin above the displayed 100% on LFP packs, so the cell genuinely has further to go before reaching its limit. Or the charge SOP map applied to LFP may simply be less restrictive near the top of charge. Both amount to more usable headroom when the car says full. Distinguishing them would require cell-level voltage measurements and the manufacturer's SOP calibration, neither of which is visible in vehicle telemetry.

Duty cycle has to be held fixed, and it is easy to get wrong

Regen capture depends far more on how a car is driven than on its state of charge. Across the LFP fleet, capture runs at about 36% of discharged energy in slow traffic and about 10% at motorway speeds. That range is several times wider than the effect being measured.

This matters because speed and state of charge are correlated in real service. Intervals measured between 95% and 100% SOC have a median speed around 24 mph, while those between 70% and 75% average about 42 mph, for the simple reason that a car which has just finished charging is usually starting a local journey. Compare the two groups without adjustment and you are largely comparing city driving against highway driving.

Left uncorrected this produces a false result, and in our case it initially appeared to show LFP capture rising at high state of charge, which is physically implausible. Coarse adjustment is not enough either: grouping all speeds above 32 mph together still hides most of the variation, because capture falls by more than twenty points inside that one group. Speed and ambient temperature are therefore treated as continuous variables, and the curves shown here are normalised across the range of duty cycles the fleet actually drives.

What this does and does not establish

Every comparison is made within a vehicle. Each car is measured against itself at different states of charge, so pack size, variant, climate, terrain and driving style are removed from the comparison rather than assumed away. Uncertainty is calculated at the vehicle level, so the confidence intervals reflect the number of cars observed rather than the number of measurements taken from them.

Three limits are worth stating plainly.

The vertical distance between the two curves is not a valid chemistry comparison. A within-vehicle design identifies the shape of each curve, not the level difference between two different populations of cars. Read each curve's own slope.

Capture is an integrated quantity. It reflects how much energy the driver sends back as well as how much the pack accepts, so it is a lower bound on what the pack would have taken had more braking energy been available.

And we did not stratify by state of health (SOH). Any interaction between pack age and the charge limit is absorbed into the per-vehicle comparison rather than resolved, so this result describes the fleet as it is rather than isolating how the effect changes as packs age.

What it means in practice

For LFP cars, charging to 100% costs nothing in regenerative braking, which is consistent with Tesla's own guidance to charge those packs fully on a regular basis.

For NCM/NCA cars the penalty is real but temporary. Capture roughly halves at a full charge, recovers as the pack drains, and is essentially gone below 80% SOC. In practice that means the opening portion of a drive rather than the drive as a whole, which is an argument for timing a 100% charge before a long journey rather than before a short local one. It is not an argument against charging to 100% when the range is needed, and it says nothing about battery health, since the energy is dissipated in the friction brakes rather than harming the pack.

This analysis comes from the battery intelligence work behind Dr.EV, our app for Tesla owners. Dr.EV covers battery health alongside the everyday features that help you manage your car. The findings reflect observed fleet behaviour, and individual results vary with climate, duty cycle and vehicle configuration.