Does LFP make more hydrogen fluoride?

A 2026 review in the Journal of Power Sources carried a line that traveled well: LFP's thermal stability is offset by severe hydrogen fluoride toxicity, at 3,000 to 8,000 ppm. Its highlights say high HF levels challenge LFP's reputation as the safest chemistry, and ESS News ran that as the headline finding. If you own a lithium iron phosphate battery, or are buying one, that deserves a straight answer.

Here it is, as best the evidence allows. HF from a failing LFP cell is a real hazard, and per kilowatt-hour LFP can make somewhat more of it than other lithium-ion types. But the gap in the available tests is a factor of two or three, not the huge gap a ppm figure suggests. ppm is a concentration, an amount divided by a volume, and LFP tops that chart largely because it vents so little of everything else.

Where the fluorine comes from

Not from the cathode. LiFePO₄ contains no fluorine. It comes mainly from the electrolyte salt, lithium hexafluorophosphate (LiPF₆), which breaks down when hot and reacts with water to give HF, and to a lesser extent from the fluorinated binder in the electrodes. All seven batteries in the 2017 study below used LiPF₆. So HF is mostly a question of how much electrolyte salt reacts, and every mainstream lithium-ion chemistry carries that salt.

Same grams, very different ppm

The clearest test I found is from Sturk and colleagues (Autoliv, RISE and Volvo Cars), published in Batteries in 2019. They heated stacks of five automotive pouch cells into thermal runaway in nitrogen, so nothing could burn, and measured what came out.

By FTIR, the LFP stack released 1.8 grams of HF and the NMC/LMO stack 1.7 grams. What differed was the rest of the gas: 50 liters from LFP against 1,500 liters from NMC/LMO. About the same HF, thirty times less gas to dilute it. The authors concluded that the HF concentration from LFP was more than an order of magnitude higher.

Their calculated composition for the LFP vent gas was 5% HF, which is 50,000 ppm. Run the same arithmetic on 1.7 grams in 1,500 liters of NMC/LMO gas and you get roughly 1,400 ppm (the conversions are mine). That is a thirtyfold gap in ppm from nearly equal grams of HF, in undiluted vent gas, and either number falls as the gas spreads. Any ppm figure depends on how far downstream you stop and measure.

Per kilowatt-hour, LFP did make more

Grams per stack is not the fair comparison, though, because LFP stores less energy. Sturk's table normalizes it: 16 g of HF per kWh for LFP against 6.0 g for NMC/LMO by FTIR, or about 2.7 times as much. Wash bottles, a second method, gave 36 and 23 g/kWh, or about 1.6 times. The authors say so plainly in their conclusions: normalized, the amount of HF is notably higher for LFP. If you are comparing two batteries of the same capacity, that is the number that matters, and it does support the review's concern in direction. It just does not support the size a raw ppm comparison implies.

Time matters too. LFP took about 45 minutes to vent; NMC/LMO took about 2 to 4. Sturk's team pointed out that once you account for that, the comparison narrows, and that which pattern is worse depends on the situation: outdoors a quick release clears fast, while a slow release in a closed space may be easier to ventilate.

Same chemistry, tenfold spread

The classic study is Larsson and colleagues in Scientific Reports (2017), which burned seven types of commercial cells with a propane burner and measured the HF per watt-hour of battery. Four of the seven were LFP, which makes it a useful check:

LFP pouch, 20 Ah: 150 to 198 mg/Wh
LFP pouch, 7 Ah: 43 to 160 mg/Wh
LFP cylindrical, 8 Ah: 52 mg/Wh
LFP cylindrical, 3.2 Ah: 12 to 24 mg/Wh
NCA-LATP pouch: 55 mg/Wh
LCO prismatic: 15 to 25 mg/Wh
Laptop pack: 15 mg/Wh

The highest and the lowest values in the study were both LFP. Three of the four LFP types reached about 50 mg/Wh or higher, level with or above every non-LFP cell tested, which fits Sturk's result, but the small LFP cylindrical cell matched the laptop pack. The authors suggest the different amounts of electrolyte and filler material may be the simple explanation, and they observe that the two LFP pouch cells gave the highest values. On this dataset the particular cell moved HF by a factor of ten, more than the chemistry label did. It is seven cell types, and no modern grid cells, so treat that as a warning against simple rankings rather than a ranking of its own.

Where the probe sits

The ppm numbers in circulation come from very different places. Three examples:

Directly above the vent. A KAUST team used a laser sensor there and reported HF orders of magnitude above the level considered immediately dangerous, for both NMC and LFP cells. I have the press release, not the paper.

In a review summary. The 3,000 to 8,000 ppm figure. It describes failure scenarios, but I could only read the abstract, so I cannot tell you which tests it comes from or how far from the cell they sampled.

In an exhaust duct. Claassen and colleagues at the Desert Research Institute burned single small cells (a 3.63 Wh LFP cell and a 9.25 Wh LCO pouch) in an 8 cubic meter chamber at about 25 air changes per hour, sampling about 2.5 meters above. Peak HF reached up to 50 ppm.

That last number looks small until you line it up against the limits. NIOSH's immediately-dangerous level for HF is 30 ppm, and the Claassen tests passed it in 36% of runs. All three LFP tests at 75% state of charge reached 44 ppm, the AEGL-3 level at which life-threatening effects are expected, and they ran higher than any other condition tested. That is one small cell, in a well-ventilated room, and the authors warn that bigger cells in an enclosed space would produce substantially higher and possibly lethal concentrations. It is why I would not read this report as reassurance.

What this means for a home battery

Grams scale with energy. Larsson's team extrapolated their results to 2 to 20 kg of HF for a 100 kWh pack and 20 to 200 kg for 1,000 kWh. Scale that to a 13.5 kWh home battery and you get about 0.3 to 2.7 kg. That is my arithmetic on their extrapolation, and it assumes the entire pack burns.

To put 270 grams in perspective: bringing it down to 30 ppm (about 25 mg per cubic meter) would take roughly 10,800 cubic meters of air, many times the volume of a house. That sum assumes perfect mixing and no losses. HF is reactive and sticks to surfaces, as Sturk's team noted, so the real number would be lower. It is not a prediction. It is a way to see that ventilation in a garage does not dilute this away, whichever chemistry is on the wall.

So what does LFP actually buy you? Plenty, just not a free pass on HF. In the KAUST tests LFP cells went into runaway at 242 to 249 °C, against 174 to 215 °C for NMC. The gap between the first small HF burst and the main one was about five minutes for LFP and one minute for NMC, which is useful warning time. Sturk measured peak stack temperatures of 386 °C against 615 °C, and a gas release rate about a hundred times lower, with thirty times less gas overall. All of that points to a slower, less violent failure. It does not make the gas that does come out any less toxic.

The practical advice is the same for any chemistry: if a battery is venting, smoking or burning, get out, stay out, and tell firefighters it is a lithium battery. Do not go look. Where the battery sits in the first place matters more than most buying decisions, which is why Report 138 goes through the placement rules. How often failures happen is a separate question, and Report 083 looks at the numbers for grid systems.

How to read an HF number

Is it ppm or grams per kWh? ppm without a location and a dilution tells you little. Mass per unit of energy can be compared.

How far from the vent? Inside the vent stream, above the cell, and in an exhaust duct are three different worlds.

Was the cell burning or just venting? Sturk's test had no flame and no oxygen. Larsson's was an external fire. Results differ.

How was HF measured? Sturk's wash-bottle masses were more than double the FTIR masses on the same gas, partly because wash bottles count all soluble fluorine and partly because HF sticks to tubing on the way to the instrument.

Is it a chemistry or a cell? Larsson's LFP cells spanned more than a tenfold range.

The honest limits on this

I read the Mrozik review's abstract and highlights only, because ScienceDirect would not open for me, so I do not know how its authors derive the 3,000 to 8,000 ppm range or how carefully the full paper handles the concentration question. My complaint is with how that line reads on its own, and the paper may well be more nuanced. The same applies to the KAUST paper, which I know only through its press release.

Sturk's test was one comparison, of one stack per chemistry, in dry, oxygen-free nitrogen. The authors say their cooling appears to have been excessive for LFP, the stack was not fully consumed, and low humidity may have limited HF formation, so a real fire could look different. So its per-kWh number could move either way in a real fire. The paper's own average-concentration figures have mixed units, so I have not used them. Larsson's cells were small batches, unused, tested in 2017 and earlier, and some values were affected by HF lost in the sampling system. None of these studies tested a complete home battery in a garage. The 13.5 kWh and 10,800 cubic meter sums are mine, and they are illustrations. For any specific system, the large-scale fire test data for that product is the thing to ask for (Report 101 explains what that test does and does not tell you).

This one is close to home for me: I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator, run by fellow veterans, and LFP is a chemistry this industry leans on. I do not own that company and earn nothing from this link. Full policy here. Nothing in this report involves them or any product they sell.