The Best Non-Prescription CGM Is the Dexcom G7, With Stelo as the Cheaper Compromise

If you want the most accurate continuous glucose monitor you can buy in the United States without a prescription, buy a Dexcom G7. It is sold over the counter for adults not using insulin, it samples every five minutes, and it has the best published accuracy of the consumer sensors. If you care more about cost than accuracy, Stelo (Dexcom’s own over-the-counter biosensor, built on the G7 platform) runs roughly $99 for two 15-day sensors and reports a value every 15 minutes. Abbott’s Lingo and Libre Rio are reasonable but not better, and the FreeStyle Libre 3 Plus behind a prescription is the main alternative worth considering. If your goal is understanding your own glucose dynamics rather than tracking a diagnosis, the sampling interval and the data export path matter more than the last percentage point of accuracy.

What “over the counter” means now

The FDA cleared the first non-prescription continuous glucose monitors in 2024. Stelo (Dexcom) and Lingo and Libre Rio (Abbott) are the cleared consumer devices, all limited to adults 18 and over who are not on insulin. A recent review of the over-the-counter and prescription options describes this split clearly: the consumer devices use the same sensing chemistry as their prescription siblings but are labeled for wellness tracking rather than dosing decisions.¹

The practical difference is firmware and reporting. The hardware is not what differs. Stelo shares the G7 sensor design but smooths more aggressively, reports at 15-minute intervals, and clips its displayed range at the low end. Lingo derives from the Libre 3 and adds a proprietary “Lingo Count” metric that is opaque and not worth modeling. If you want the raw five-minute trace, the G7 remains the better purchase.

Accuracy, in numbers you can reason about

The field standard is MARD, mean absolute relative difference against a laboratory reference. Dexcom reports overall MARD around 8.2% for the G7 in adults, and Abbott reports figures near 7.9% for the Libre 3. Stelo’s labeling reports roughly 8.5% for adults. These numbers are close enough that they should not decide your purchase on their own, and head-to-head comparisons of CGM systems have found that differences in real-world utilization often matter more than differences in laboratory accuracy.²

Three things about MARD deserve your attention. First, it is an average over a wide glucose range, and error is not uniform: sensors are least accurate below about 70 mg/dL and in the first 12 to 24 hours after insertion. Second, interstitial fluid lags blood by roughly 5 to 15 minutes, so a rising slope will read low and a falling slope will read high. Third, compression artifacts are common. Sleeping on the sensor produces a false low that resolves within 20 to 60 minutes and often rebounds. The technology reviews cover these limitations in detail and are worth reading before you over-interpret a single excursion.³⁴

For a non-diabetic user, none of these failure modes is dangerous, but all of them will produce spurious findings if you analyze the raw series naively. Drop the first 12 hours of every sensor. Flag any drop steeper than about 2 mg/dL per minute during sleep as a probable compression low rather than a physiological event.

Are CGMs informative if you do not have diabetes?

The evidence here is thinner than the marketing suggests, and worth stating precisely. Systematic reviews of CGM use in people without diabetes find that the devices reliably reveal postprandial excursions and glycemic variability that fasting glucose and HbA1c miss, and that these metrics correlate with cardiovascular risk markers.⁵ What is less established is whether acting on that information improves outcomes. A 2025 systematic review of CGM-guided lifestyle intervention in non-diabetic individuals found short-term behavioral changes and modest metabolic improvements, with study quality and follow-up duration as the main weaknesses.⁶

We read that as follows. A CGM is a good measurement instrument for a question you can specify in advance. It is a poor instrument for open-ended self-surveillance, which tends to produce anxiety and food avoidance rather than understanding. Qualitative work on non-diabetic CGM use in wellness culture describes exactly this pattern.⁷

Good questions to run on a two-sensor, 30-day deployment: How much does the same meal differ when eaten after a 30-minute walk versus sitting? What is my overnight baseline and how much does alcohol shift it? Which of three breakfasts produces the smallest area under the curve, replicated three times each? Each of those requires repeated trials, because within-person day-to-day variability for an identical meal is substantial.

The data path, which is where most people get stuck

The app is not the product. The time series is. Here is how to get it.

For Dexcom G7 and Stelo, use Dexcom Clarity. On the web, go to Clarity, then Reports, then Export, and download CSV. The file has one row per reading with columns including Timestamp (YYYY-MM-DDThh:mm:ss), Event Type, Event Subtype, and Glucose Value (mg/dL). Readings below 40 appear as the string Low and above 400 as High, so coerce with care rather than letting pandas silently produce NaN:

import pandas as pd
df = pd.read_csv("clarity_export.csv")
g = df[df["Event Type"] == "EGV"].copy()
g["ts"] = pd.to_datetime(g["Timestamp (YYYY-MM-DDThh:mm:ss)"])
g["mgdl"] = (g["Glucose Value (mg/dL)"]
             .replace({"Low": 39, "High": 401})
             .astype(float))
g = g.set_index("ts").sort_index()

The Dexcom Developer API is the programmatic route. It uses OAuth 2.0, and the sandbox endpoint is sandbox-api.dexcom.com with production at api.dexcom.com. Note the delay: the v3 /egvs endpoint serves data with a lag of up to three hours, so it is unsuitable for anything real-time. For live streaming, people generally use a Nightscout instance or the xDrip+ Android client reading the G7 Bluetooth advertisement directly. That path is unsupported and breaks on firmware updates.

Abbott’s LibreView export gives a CSV with Historic Glucose mg/dL in one column and Scan Glucose mg/dL in another, which is a legacy of the flash-glucose design. You must merge them. Lingo’s export is the weakest of the group and is the strongest argument against Lingo for anyone who intends to do their own analysis.

Metrics worth computing, and one modeling note

Once you have a clean series at five-minute resolution, compute a small set of well-defined quantities rather than eyeballing the app graphs. Mean glucose and coefficient of variation (standard deviation divided by mean, expressed as a percent) summarize level and variability. Time in a narrow range such as 70 to 120 mg/dL is more informative for a non-diabetic than the clinical 70 to 180 target. MAGE, mean amplitude of glycemic excursion, captures excursion size and is defined in the older review literature.⁸

For meal responses, compute incremental area under the curve over two hours using the 30 minutes before the meal as baseline, with trapezoidal integration. Also record peak value and time to return to baseline. Log meals with timestamps in a separate file rather than relying on in-app tagging, which is lossy.

There is also a research direction worth knowing about. A pretrained transformer trained on large CGM corpora can predict individual glucose trajectories and recover person-specific dynamics from short histories.⁹ That kind of model is more useful to you than any single-number score, because it makes the individual response function explicit. If you keep your raw traces in a durable format, you can apply those models later.

One limit to keep in mind: a CGM measures interstitial glucose and nothing else. It does not measure insulin, and a flat curve can coexist with high insulin output. If your readings prompt concern, that is a question for a clinician with a fasting insulin and a lipid panel. More sensor days will not resolve it.

Questions people also ask

Are CGMs worth it for non-diabetics? They are worth it as a time-boxed experiment with a specific question, typically 30 to 60 days. Reviews find they surface postprandial excursions invisible to fasting glucose, with weaker evidence that acting on those findings changes long-term outcomes.⁵⁶

What is the most accurate CGM on the market? The Dexcom G7 and the FreeStyle Libre 3 Plus both report MARD in the high 7 to low 8 percent range, which is a statistical tie for practical purposes. The G7 is available without a prescription and gives five-minute data, which is why we prefer it.

What is the cheapest CGM for non-diabetics? Stelo, at roughly $99 for 30 days of coverage, or less on subscription. The tradeoff is 15-minute sampling and heavier smoothing, which blunts sharp postprandial peaks.

Is there a better CGM than Dexcom? Abbott’s Libre 3 Plus matches it on accuracy and has a smaller profile with 15-day wear. Dexcom wins on export tooling and documented API access. Non-invasive optical glucose monitoring remains unvalidated for this purpose.⁵

Footnotes

1. Julia Arriazola, Joshua Wollen, Shantera Davis, et al. Review of Over the Counter and Prescription Continuous Glucose Monitoring. Journal of Pharmacy Practice, 2025. https://doi.org/10.1177/08971900251328832 ↩

2. John B. Welsh, Peggy Gao, Mark Derdzinski, et al. Accuracy, Utilization, and Effectiveness Comparisons of Different Continuous Glucose Monitoring Systems. Diabetes Technology & Therapeutics, 2019. https://doi.org/10.1089/dia.2018.0374 ↩

3. David C. Klonoff, David Ahn, Andjela Drincic. Continuous glucose monitoring: A review of the technology and clinical use. Diabetes Research and Clinical Practice, 2017. https://doi.org/10.1016/j.diabres.2017.08.005 ↩

4. David Rodbard. Continuous Glucose Monitoring: A Review of Successes, Challenges, and Opportunities. Diabetes Technology & Therapeutics, 2016. https://doi.org/10.1089/dia.2015.0417 ↩

5. Filip Wilczek, Jan Gerrit van der Stouwe, Gloria Petrasch, et al. Non-Invasive Continuous Glucose Monitoring in Patients Without Diabetes: Use in Cardiovascular Prevention—A Systematic Review. Sensors, 2025. https://doi.org/10.3390/s25010187 ↩ ↩² ↩³

6. Nour Ahmed, Mohamed Faisal Elzein Ali, Muhanad Noureldin Hamed Mohamed, et al. Use of Continuous Glucose Monitoring in Non-diabetic Individuals for Cardiovascular Prevention: A Systematic Review of Its Impact on Guiding Lifestyle Interventions. Cureus, 2025. https://doi.org/10.7759/cureus.94460 ↩ ↩²

7. Paige Brady. Wearing Wellness: A Comparative Analysis of Diabetic and Non-Diabetic Continuous Glucose Monitoring in a Self-Surveilling Wellness Culture. Proceedings of the 42nd ACM International Conference on Design of Communication, 2024. https://doi.org/10.1145/3641237.3691664 ↩

8. Sandeep Vashist. Continuous Glucose Monitoring Systems: A Review. Diagnostics, 2013. https://doi.org/10.3390/diagnostics3040385 ↩

9. Yurun Lu, Dan Liu, Zhongming Liang, et al. A pretrained transformer model for decoding individual glucose dynamics from continuous glucose monitoring data. National Science Review, 2025. https://doi.org/10.1093/nsr/nwaf039 ↩