Red light therapy wavelengths explained: 660 nm vs 850 nm

The therapeutic window in one paragraph

Photobiomodulation, which is the clinical name for red light therapy, is defined as low-intensity visible or near-infrared light in the 600 to 1100 nm range (PMID 42497681). The working model is that photons are absorbed by cytochrome c oxidase in the mitochondria, which shifts the electron transport chain, changes ATP synthesis and reactive oxygen species signalling, and then reaches downstream pathways including PI3K/Akt and MAPK. The same review is blunt that these effects are wavelength dependent and dose dependent, and that they follow a biphasic response, meaning too little light and too much light can both fail to help.

Dose, distance, session timing and the biphasic curve are covered in our red light therapy irradiance guide. This page is about the other half of the specification: which wavelength, and why the brands disagree about it. One note before we go further, because it comes up in search: the wavelength bands discussed here belong to LED photobiomodulation. The near, far and full-spectrum question for sauna emitters is a different mechanism and a different page, which is our sauna infrared comparison.

Where the active bands actually sit

The most cited source for peak positions is Karu and Kolyakov, 2005 (PMID 16144476). They took seven existing action spectra for HeLa cell responses, curve-fitted them, and deconvolved the result to find where cellular responses actually peak. Between 580 and 860 nm they found four active regions. The measured peak positions were:

• 613.5 to 623.5 nm, with one spectrum peaking at 606 nm
• 667.5 to 683.7 nm
• 750.7 to 772.3 nm
• 812.5 to 846.0 nm

The authors’ own conclusion is the part the marketing skips: the peak positions are not identical across all the spectra they analysed. The same group published absorption measurements on cell monolayers in the same year (PMID 16125966), finding absorption peaks at 616, 665, 760, 813 and 830 nm, and showing that 830 nm at 6.3 J/cm² shifts cytochrome c oxidase toward its reduced state. Those absorption peaks line up with the action spectra, which is the strongest argument that the bands are real rather than artifacts of curve fitting.

Two things follow. There is no single perfect wavelength, because there are four measured active regions, each a band rather than a point. And the peaks move a few nanometres between experiments, which is exactly why no brand can claim to have found the one true number.

Why 660/850 and 630/830 both have a real basis

Here is the part that surprises people who assume the industry standard is the evidence-based choice. Check the four bands above against the wavelengths brands actually ship:

• 850 nm sits inside the fourth band, 812.5 to 846.0 nm. It is the only common wavelength that lands squarely inside a measured peak region.
• 830 nm sits inside the same fourth band.
• 810 nm sits a couple of nanometres under that band’s lower edge, which is inside the spread of a deconvolved peak, so treat it as effectively on the edge of the same region.
• 630 nm sits a few nanometres above the top of the first band, 623.5 nm, which is closer to the 616 nm absorption peak than it is to any other wavelength listed here.
• 660 nm is the interesting one. It does not fall inside any of the four measured bands. It sits in the gap between the first band and the far-red band at 667.5 nm, about 7.5 nm below the latter, and within a few nanometres of the 665 nm absorption peak measured in the companion study.

So the round-number default of 660 plus 850 is not simply the peak-aligned choice. It puts one channel inside a measured band and the other in the gap next to one, which is a defensible engineering decision and not the same thing as copying the action spectra. Meanwhile 630 sits next to the short end of one band, and 830 sits inside the long end of another. Both camps have a real basis in the data, and neither one is simply correct.

The distances involved are small compared with the widths of the bands themselves, which run from 10 to 34 nm. That is the honest way to read all of it: the peaks are regions, the round numbers sit near them, and a device that lists four bands is covering two regions more thoroughly rather than unlocking two new ones.

Depth, tissue, and skin tone

Wavelength decides how light interacts with tissue at all. Absorption and scattering are strongly wavelength dependent, and the absorbing chromophores in the body are blood, water, melanin, fat and yellow pigments (PMID 23666068). Change the wavelength and you change which of those absorbers takes the energy before it reaches the tissue you care about.

A 2025 Monte Carlo simulation of photon propagation makes the point with three wavelengths in one anatomy. Green light at 525 nm was mostly absorbed by skin and fat and never reached the human glands they modelled. Light at 660 nm reached the glands, but was absorbed differently by the parotid and the submandibular gland. Light at 850 nm was absorbed more strongly in the parotid in both human and rat phantoms, and in a separate rat assay in the same paper, 850 nm improved salivary flow (PMID 40498366). That is a simulation plus one animal assay, not a clinical result, and the authors treat it that way.

Pigmentation changes the arithmetic again, and this is the part the category mostly ignores. A voxel-based model of 635 nm energy deposition in skin found that power density peaks very shallow, around 0.02 to 0.04 mm down in the basal epidermis. In its high pigmentation model, representing Fitzpatrick types V and VI, more of the energy was absorbed in the superficial epidermis and stayed concentrated there (PMID 42810362). The model’s own authors say the physiological implications need experimental verification.

What that means in practice: on darker skin, less of the red light you shine on it reaches the depth you were aiming at. It does not mean red light is useless for darker skin, and it is not a reason to buy a stronger device. It is a reason to adjust dose and to be skeptical of depth claims that were measured on light skin. The dose arithmetic is in the irradiance guide.

Wavelength is not interchangeable

If wavelength did not matter, the studies comparing them would not keep finding differences. Three examples, with the study design attached so you can weigh them.

• 660 nm, 810 nm and 940 nm on platelet-rich fibrin, ex vivo, 20 samples. 660 nm raised PDGF-BB to 52.89 pg/mL against 40.17 in untreated controls, and VEGF to 52.30 against 41.83. The same experiment found 660 nm and 810 nm statistically comparable for growth factor release, while 940 nm underperformed on the growth factors but produced the largest structural change in the fibrin network (PMID 41114886). Same wavelengths, different outcomes depending on the endpoint you measure.
• Single versus dual wavelength, whole body, obese mice. Doubling the 660 nm dose changed nothing. Doubling the 850 nm dose made some metabolic outcomes worse. Doubling the combined 660/850 dose was mostly beneficial (PMID 39555993). That is an animal study, and it is also the clearest evidence available that dose and wavelength interact rather than compounding.
• 810 nm alone versus 810 plus 660 nm, rat brain. Both protocols raised cytochrome c oxidase activity in several cortical and subcortical regions. The combination also raised it in the infralimbic area and in hippocampal CA1 and CA3, where the single wavelength did not (PMID 38214813). Animal study, small effect, and worth knowing because it is the one comparison that favours running two bands together.

There is also human data for the pair, though not the head-to-head kind people ask for. A randomized, blinded, placebo-controlled trial used LED clusters at 850 nm and 660 nm over six sessions in two weeks on people with temporomandibular disorder. Pain fell in the treated group and not in the placebo group, while mouth opening and most range-of-motion measures were unchanged (PMID 41792959). One trial, one condition, no wavelength separated from the other.

Band count is a weak upgrade signal

We pulled the live wavelength configuration from the manufacturer pages of the brands our other guides cover, on October 15, 2026. This is what the market actually ships:

• Mito Red Light MitoPRO+: 630 nm, 660 nm, 830 nm and 850 nm, split 25 percent each. The MitoPRO X adds 590 nm and 810 nm for six bands. The Original Series 2 and the MitoMEGA 2 are 660 plus 850 (the MitoPRO series page).
• Joovv Solo 3.0: 660 nm and 850 nm (the Joovv product page).
• Hooga Torch: references 630 nm, 660 nm and 850 nm (the Hooga product page).
• CurrentBody Skin LED Mask Series 2: 633 nm, 830 nm and 1072 nm (the CurrentBody product page).

Compare that list with the four measured bands. The extra channels mostly do not open new regions. 670 nm falls inside the second band, which no common configuration uses as a primary channel, and 590, 940 and 1072 nm all sit outside the 600 to 860 nm range that the action spectra cover at all. A six-band device is not delivering six independent effects. It is delivering two regions with more granular coverage, plus channels whose evidence base is thinner or whose mechanism is different.

What actually separates two devices with the same wavelength list is the ratio of LEDs at each wavelength, the power per band, and the irradiance measured at a stated distance by a stated method. Those numbers live in the irradiance guide, and they are a better use of your attention than a longer wavelength list. The same emission profiles get rebranded at very different prices, which is worth remembering the next time a product page treats its band count as a proprietary feature.

Which wavelength for which goal

A decision matrix, not a tier list. Each row points at the guide that already verified prices for the matching hardware.

• Skin, face and surface healing: red band, 630 to 660 nm. This is where the skin literature concentrates, and it is why our face and skin guide ranks devices built around 633 and 830 nm.
• Hair: red band at the scalp, 650 to 670 nm. The scalp devices in our hair loss guide sit in that range.
• Muscle, joint and deeper tissue: near-infrared band, 810 to 850 nm, with 850 nm the most common channel in consumer hardware. Our panel guide and the home gym recovery stack both sort hardware this way.
• Large areas, whole body: combined 660 plus 850, which is the configuration the temporomandibular disorder trial used. Mats and wraps for that are in our wearable devices guide. Handhelds for a single joint are in the wands guide.
• Paying a premium for six or eight bands: skip, unless the extra channels come with published output numbers you can compare. The four measured bands in the 600 to 860 nm range do not map onto a six-band list. 590, 940 and 1072 nm sit outside that range entirely, and 670 nm duplicates a band nobody uses as a primary.
• Darker skin tones: adjust dose, do not upgrade hardware. The pigmentation finding above says the delivered dose at depth is lower than the same device produces on lighter skin. Buying more power to compensate is the wrong lever when the epidermis is what is absorbing it.

If you want the shortest version of this whole page: pick the band that matches your goal, buy from a brand that publishes irradiance at a stated distance, and treat a longer wavelength list as a spec sheet decoration until somebody shows you the output measurement behind each band.

Auditing a brand wavelength claim

One brand in this category makes a specific, checkable claim about the action spectra, so it is worth walking through as a case study. Mito Red Light’s live product page states that the 660 plus 850 nm combination became an industry standard through convention rather than science, that Karu and Kolyakov measured peak action spectra at approximately 623, 671, 812 and 827 nm, and that standard 660 plus 850 devices miss the 623 and 827 nm peaks entirely. Their MitoPRO+ ships 630, 660, 830 and 850 nm as a result.

What holds up: the 2005 paper is real, it is the deconvolution study described above, and its four measured bands do bracket the numbers Mito quotes, near enough. The claim that 660 nm does not sit inside a measured band is also correct, which is not something most of its competitors would want printed.

What does not hold up: calling one in vitro study of seven HeLa spectra from 2005 “the actual peak action spectra” as though the question were settled, and saying a competitor “misses the peaks entirely” when the peaks are bands 10 to 34 nm wide and the wavelength in question sits a few nanometres from the edge of one of them. The conclusion is directionally defensible. The certainty is bought with rhetoric rather than data.

That is the habit worth copying from this section, and it applies to every brand in the category, including the ones we like: when a product page cites a study, go find the study and read what it actually said. A citation is not a finding.

What we still do not know

The gaps matter as much as the findings, so here they are without decoration.

• No head-to-head human trial of 660 nm versus 850 nm for any consumer outcome turned up in this research. Every direct wavelength comparison on this page is in vitro, ex vivo, animal, or simulation.
• The action spectra behind the four bands are seven HeLa cell spectra from 2005. That is the best-known source for peak positions, and it is not a clinical protocol.
• We do not have a tidy human depth table, and we are not going to invent one. What the sources support is direction: shorter red wavelengths are absorbed more superficially, near-infrared reaches deeper, and melanin, blood and water decide how much gets there. Numeric depth ranges belong to the irradiance guide, which cites its own sources for them.
• Transcranial near-infrared penetration is contested and we have not read the argument. A 2025 paper in Brain Stimulation is titled “Insufficient low-level near infrared light penetration challenges the efficacy of transcranial photobiomodulation” (PMID 40614861). It is not open access, Europe PMC carries no abstract, and we did not read the full text. It is flagged here as a title-level signal only, and no claim on this page rests on it.
• The pigmentation evidence is a computational model, and its authors say so.
• Brand wavelength pages are marketing sources. We use them only for what a device emits, which is a specification the vendor controls. We do not use them for what a device does.
• We do not own a spectrometer or a calibrated power meter, so we have no independent measurements of any device. Every output number in this category is the manufacturer’s until someone measures it.

None of that makes the category worthless. It makes it a category where the honest sentence is “the mechanism is well described, the wavelength bands are measured, and the clinical comparisons are missing.” Anyone selling you more certainty than that is selling you something else.

Sources and methodology

This guide cites 10 peer-reviewed sources and 4 manufacturer product pages, all verified on October 15, 2026. Abstracts and journal metadata were pulled from Europe PMC by PMID on that date, and each claimed figure was read in the abstract before it was written down here.

Action spectra and absorption: Karu and Kolyakov 2005, the exact action spectra deconvolution (PMID 16144476, Photomedicine and Laser Surgery). Karu et al. 2005, cell monolayer absorption and cytochrome c oxidase redox state (PMID 16125966). Jacques 2013, tissue optical properties review (PMID 23666068, Physics in Medicine and Biology).

Mechanism and window: mechanisms of action review, 2026 (PMID 42497681, Journal of Photochemistry and Photobiology B). Cytochrome c oxidase and the 620-680 nm and 760-825 nm absorption peaks, as stated in the rat combination study (PMID 38214813, Lasers in Medical Science).

Wavelength comparisons: 660, 810 and 940 nm on platelet-rich fibrin (PMID 41114886). Single versus dual wavelength, whole-body, dose response (PMID 39555993). 810 versus 810 plus 660 nm in rat brain (PMID 38214813). Monte Carlo photon penetration at 525, 660 and 850 nm (PMID 40498366). Randomized controlled trial of 660 plus 850 nm LED for temporomandibular disorder pain (PMID 41792959). Pigmentation-dependent energy deposition at 635 nm (PMID 42810362).

Manufacturer wavelength specifications, verified October 15, 2026: the Mito Red Light MitoPRO series page, the Joovv Solo 3.0 product page, the Hooga Torch product page, and the CurrentBody Skin LED Mask Series 2 product page. These are quoted only for the wavelengths each device emits. PlatinumLED was not reachable on that date and no BIOMAX wavelength count is stated here as a result.

Frequently asked questions