Over-the-Counter CGMs: What You Can Buy, What They Cost, and How Accurate They Are

Yes, you can buy a continuous glucose monitor in the United States without a prescription. Since March 2024, the FDA has cleared three over-the-counter integrated CGMs: Dexcom’s Stelo, Abbott’s Lingo, and Abbott’s Libre Rio. All three are intended for adults who do not use insulin. Pricing runs roughly $89 to $99 for a two-sensor box covering 28 to 30 days, or about $79 to $89 per month on subscription, which puts a year of continuous glucose data in the $950 to $1,200 range with no insurance involved. Accuracy is expressed as mean absolute relative difference (MARD) against a laboratory reference. It sits in the 8 to 10 percent band for the underlying sensor platforms, which is good enough for pattern-finding and bad enough that you should not treat any single reading as ground truth.

The three OTC devices and how they differ

Stelo is built on the Dexcom G7 sensor hardware, worn on the back of the upper arm, and runs 15 days per sensor. Lingo and Libre Rio are both Abbott products built on the FreeStyle Libre 3 lineage, with Lingo running 14 days and marketed toward metabolic self-experimentation, and Libre Rio running 14 days and aimed at adults with type 2 diabetes not on insulin. The FDA’s OTC clearances were a genuine shift in access, moving CGM out of a prescription-gated diabetes device category and into a consumer wearable category ¹.

The differences that matter for a technical user are not in the sensor chemistry, which is broadly the same across the family: an enzymatic electrode using glucose oxidase or glucose dehydrogenase on a filament inserted about 5 mm into the subcutaneous interstitial fluid, producing a current proportional to glucose concentration ². The differences are in reporting range, smoothing, and data access. Stelo reports a range of roughly 70 to 250 mg/dL and does not alarm for hypoglycemia. Lingo displays a proprietary “Lingo Count” derived from glucose excursions alongside the mg/dL trace. Libre Rio reports 40 to 400 mg/dL, the widest of the three.

We would choose Stelo if the goal is a clean glucose trace with the least editorializing, and Libre Rio if you want the full physiological range and expect excursions outside 70 to 250. Lingo’s coaching layer is fine but adds an abstraction between you and the numbers that you will want to strip off anyway.

What “accurate” means here, precisely

MARD is the mean of the absolute relative difference between each sensor reading and a paired reference measurement, usually venous plasma glucose run on a YSI analyzer. A MARD of 9 percent means that on average a sensor reading at a true 100 mg/dL lands about 9 mg/dL off, in either direction. The distribution has tails: a small percentage of readings are off by 20 percent or more, and those errors cluster in specific conditions.

Three failure modes dominate, and you should learn to recognize all three in your own traces. The first is the interstitial lag. CGM measures glucose in interstitial fluid, not blood, and the equilibration delay is typically 5 to 15 minutes. During a fast rise after a high-carbohydrate meal, the sensor reads low relative to blood. During a rapid fall, it reads high. This lag is the single largest contributor to apparent inaccuracy during dynamic periods, and it is physiology rather than sensor defect ³.

The second is compression artifact, sometimes called a compression low. Lying on the sensor reduces local perfusion and produces a sharp and physiologically implausible drop, often 30 to 50 mg/dL over 20 minutes. The reading recovers just as sharply when you roll over. These are trivially identifiable in overnight data by their squared-off shape and their correlation with sleep position. Do not include them in nocturnal minimum calculations.

The third is chemical interference. Electrochemical CGMs respond to substances other than glucose that are oxidizable at the working electrode potential or that interact with the enzyme layer. Ascorbic acid (vitamin C) at high supplemental doses, acetaminophen on some platforms, and hydroxyurea are the well-characterized ones, and the mechanism differs by sensor generation and membrane design ⁴. In vivo, the clinical significance varies with dose, timing, and the specific device, and the field still lacks a standardized interference testing protocol across manufacturers ⁵. If you take gram-scale vitamin C, expect positive bias on some sensors and annotate accordingly.

The first 12 to 24 hours after insertion are also the least reliable, as the wound-healing response around the filament settles. We discard day one from any analysis.

Cost accounted

Sticker price is $89 to $99 per 28- to 30-day box, or about $79 to $89 monthly on subscription. Budget for sensor loss: adhesive failure and accidental knocks cost most people one sensor in ten to fifteen, so the realistic annual figure is closer to $1,100 than $950. Overpatches (Skin Grip, Simpatch, or similar) run about $0.50 each and cut that failure rate substantially. There is no meter, no lancets, and no strips.

If you want to validate the sensor against blood, add a standard fingerstick meter and strips, roughly $20 plus $0.30 per strip. We recommend running ten paired fingersticks per sensor, spread across fasting, post-meal peak, and pre-bed, and computing your own MARD. It takes fifteen minutes of effort per sensor and tells you whether a given sensor is running with a systematic bias, which happens often enough to matter.

Getting the raw data out

The manufacturer apps are the default sink and the weakest link. All three platforms let you export a CSV, and the export is the file you should build on.

From Abbott’s LibreView web portal, the export gives a per-row structure with Device Timestamp, Record Type (0 for historic 15-minute readings, 1 for scan readings, 5 for notes), and Historic Glucose mg/dL. Filter to Record Type == 0 for an evenly spaced series. Stelo exports from the Dexcom Clarity web interface with Timestamp (YYYY-MM-DDThh:mm:ss), Event Type, and Glucose Value (mg/dL), where values outside range appear as the strings Low and High rather than numbers. Coerce those explicitly before any arithmetic. Silently dropping them biases your variability metrics downward.

A minimal pandas pipeline we use:

import pandas as pd
df = pd.read_csv("clarity_export.csv", skiprows=1)
df = df[df["Event Type"] == "EGV"].copy()
df["ts"] = pd.to_datetime(df["Timestamp (YYYY-MM-DDThh:mm:ss)"])
df["glucose"] = pd.to_numeric(
    df["Glucose Value (mg/dL)"].replace({"Low": 39, "High": 401}),
    errors="coerce")
df = df.set_index("ts").resample("5min").mean().interpolate(limit=3)

The derived metrics worth computing are time in range (percent of readings 70 to 180 mg/dL), coefficient of variation (standard deviation divided by mean, with under 36 percent the conventional stability threshold), mean amplitude of glycemic excursion, and the glucose management indicator. These definitions come out of the diabetes literature and carry specific clinical meaning in that population ⁶. Applied to a person without diabetes they are descriptive statistics about your own variability. They are not a diagnostic output.

What CGM data does and does not tell you if you do not have diabetes

This deserves plain language. The evidence base for CGM in people without diabetes is thin, and the observed glucose variability in healthy adults is largely normal physiology rather than a signal of disease ⁷. Postprandial excursions to 140 or 160 mg/dL after a large carbohydrate load are common in people with entirely normal fasting glucose and HbA1c. Reading those peaks as damage is unsupported.

What CGM is good at is within-person comparison under controlled conditions: the same meal eaten at 8 a.m. versus 8 p.m., the same meal with and without a 15-minute walk after, the effect of a night of short sleep on the next morning’s fasting level. Hold everything constant except one variable, repeat it three or more times, and you have a usable personal experiment. Evidence for CGM’s value outside intensive insulin therapy is accumulating but still uneven, and access and interpretation remain the open problems ⁸.

If your data shows a sustained fasting glucose above roughly 100 mg/dL across multiple sensors, or peaks repeatedly above 200 mg/dL, that is a reason to see a clinician for proper laboratory testing. That testing means fasting plasma glucose, HbA1c and an oral glucose tolerance test. A CGM trace is not a diagnosis, and no OTC device is cleared to make one.

Questions people also ask

Can you get a CGM over the counter, and do you need a prescription?

In the US, no prescription is required for Stelo, Lingo, or Libre Rio. You buy them online or in pharmacies the way you buy any consumer wearable. Prescription CGMs (Dexcom G7, FreeStyle Libre 3 Plus) still require a clinician and are the right choice if you use insulin, because the OTC versions lack hypoglycemia alarms and insulin-dosing indications. Regulatory status differs outside the US, and European availability follows a separate pathway ⁹.

What is the cheapest CGM without insurance?

The OTC devices are the cheapest route, at roughly $79 to $99 per month depending on whether you subscribe or buy boxes outright. That is a fraction of the historical cash price for prescription CGM, which ran $200 to $300 per month before pharmacy benefits.

Are over-the-counter CGMs accurate?

The sensor platforms report MARD in the 8 to 10 percent range against laboratory reference, which is comparable to their prescription counterparts since the hardware is shared. Accuracy degrades during rapid glucose change, under physical compression, in the first day of wear, and in the presence of interfering compounds such as high-dose ascorbic acid ⁴.

Which OTC CGM is most accurate?

Published head-to-head comparisons of the three OTC products specifically are limited, and the underlying platforms perform similarly. We would select on reporting range and data access rather than on claimed MARD: Libre Rio for its 40 to 400 mg/dL range, Stelo for the least-processed trace.

How long does a sensor last, and what happens when it fails?

Stelo runs 15 days, the Abbott devices 14. Early failure from adhesive loss or filament dislodgement happens in roughly one sensor in ten to fifteen. Both manufacturers replace failed sensors on request, though the process costs you days of data continuity, so keep a spare on hand if you are running a timed experiment.

Footnotes

1. Zain Afridi, Sameer Abdul Rauf, Syed Muhammad Nabil Ashraf, et al. Transformative Advances in Continuous Glucose Monitoring and the Impact of FDA Over‐the‐Counter Approval on Diabetes Care. Health Science Reports, 2025. https://doi.org/10.1002/hsr2.70747 ↩

2. Raeann Gifford. Continuous Glucose Monitoring: 40 Years, What We′ve Learned and What’s Next. ChemPhysChem, 2013. https://doi.org/10.1002/cphc.201300172 ↩

3. So Yoon Kwon, Jun Sung Moon. Advances in Continuous Glucose Monitoring: Clinical Applications. Endocrinology and Metabolism, 2025. https://doi.org/10.3803/enm.2025.2370 ↩

4. Steven John Setford. The Impact of Interfering Substances on Continuous Glucose Monitors: Part 1: Classification of Continuous Glucose Monitoring Devices and Mechanisms of Substance Interference. Journal of Diabetes Science and Technology, 2025. https://doi.org/10.1177/19322968251377027 ↩ ↩²

5. Steven John Setford. The Impact of Interfering Substances on Continuous Glucose Monitors: Part 4: Continuous Glucose Monitoring Interferents: In Vivo Performance, Clinical Implications and Future Standardization. Journal of Diabetes Science and Technology, 2025. https://doi.org/10.1177/19322968251377026 ↩

6. Thomas C. Blevins, Bruce W. Bode, Satish K. Garg, et al. Statement by the American Association of Clinical Endocrinologists Consensus Panel on Continuous Glucose Monitoring. Endocrine Practice, 2010. https://doi.org/10.4158/ep.16.5.730 ↩

7. Minna Johansson, Justin A. Dower, Kasia J. Lipska. Continuous Glucose Monitoring for Persons With and Without Diabetes. JAMA Internal Medicine, 2026. https://doi.org/10.1001/jamainternmed.2026.2769 ↩

8. Andrzej Bazan, Oliwer Plichta, Natalia Radzinska, et al. Continuous Glucose Monitoring Beyond Intensive Insulin Therapy in Type 2 Diabetes: Evidence, Over-the-Counter Devices, Equity, and Clinical Implementation. Cureus, 2026. https://doi.org/10.7759/cureus.112200 ↩

9. Oskar Kublin, Dorota Filipowicz, Mariusz Stępień. Continuous glucose monitoring (CGM) systems in Poland, Europe, and beyond in 2025: principles, available technologies, and future perspectives. Archives of Medical Science, 2026. https://doi.org/10.5114/aoms/218227 ↩