Metformin has been prescribed since the 1950s, costs a rounding error next to a weekly injection, and almost never appears on a telehealth weight page. Part of that is commercial and part of it is fair: the weight effect is small, and small is what the category is selling against. But the trial record on metformin runs fifteen years, which is roughly fourteen years longer than the record behind the figures in the semaglutide article, and the comparison changes shape depending on where along that line it is taken.
What metformin was actually tested to do
The Diabetes Prevention Program randomly assigned 3,234 adults with elevated fasting and post-load glucose to placebo, to metformin 850 mg twice daily, or to a lifestyle program targeting a 7% weight loss and 150 minutes of weekly activity. Mean age was 51, mean body-mass index 34.0. Over an average 2.8 years, diabetes incidence was 11.0, 7.8 and 4.8 cases per 100 person-years across the three arms.[1]
That is a 31% reduction on metformin (95% CI, 17% to 43%) against 58% on lifestyle (95% CI, 48% to 66%), and the lifestyle program beat the drug significantly. To prevent one case of diabetes over three years, 13.9 people had to take metformin and 6.9 had to complete the lifestyle program.[1] Weight was never the primary endpoint. Diabetes was.
The prevention effect did not evaporate, and it did not stay ranked either. Followed to fifteen years, diabetes incidence remained reduced by 18% on metformin (HR 0.82; 95% CI, 0.72 to 0.93) and 27% on lifestyle (HR 0.73; 95% CI, 0.65 to 0.83). By year 15 the cumulative incidence of diabetes was 55% in the lifestyle group and 56% in the metformin group, against 62% on placebo — the two active arms had converged. No arm showed a significant reduction in the aggregate microvascular outcome.[2]
The fifteen-year reversal
A later analysis followed the 1,066 participants who had lost at least 5% of their weight in the first year. At one year the ranking was the obvious one: 62.6% of the lifestyle group had reached that threshold, against 28.5% on metformin and 13.4% on placebo.[3]
Then the order inverts. Across years six to fifteen, mean weight loss maintained relative to baseline was 6.2% in the metformin group (95% CI, 5.2% to 7.2%), 3.7% in the lifestyle group (95% CI, 3.1% to 4.4%) and 2.8% on placebo. Continued metformin use was itself an independent predictor of holding the loss.[3] The arm that induced the least kept the most.
The caveat belongs in the same breath: this is a post-hoc analysis of non-randomized subsets of randomized groups, and it describes only people in whom the treatment had already worked once. It is not evidence that metformin outperforms anything in an unselected population. It is evidence that durability and initial magnitude are separate properties, which is the question the withdrawal trials ask of the newer class.
The size of the gap, stated plainly
Nothing above makes metformin competitive on magnitude. STEP 1 put semaglutide 2.4 mg at a mean −14.9% of body weight over 68 weeks against −2.4% on placebo.[4] SURMOUNT-1 put tirzepatide 15 mg at −20.9% over 72 weeks against −3.1%.[5] Against a metformin arm whose maintained loss was 6.2% in its best-responding subgroup, that is a different order of effect, and a reader choosing on first-year weight is not choosing wrongly.
Where the two have been compared directly
No randomized trial has assigned adults with obesity to metformin or to a GLP-1 with weight as the endpoint. The nearest randomized evidence asks an add-on question instead. SURPASS-EARLY randomized 794 adults with up to four years of type 2 diabetes, all already on metformin, to tirzepatide or to intensified conventional care. At two years, glycated hemoglobin fell 1.99 points against 1.32, an estimated treatment difference of −0.68 points (95% CI, −0.84 to −0.51; P < 0.001), with a weight difference of −8.0 kg (95% CI, −9.39 to −6.50). Normoglycemia was reached by 60.2% against 24.0%.[6]
A direct comparison exists only in observational data, and in a narrow population. A matched cohort of 3,115 pairs of antipsychotic-treated adults with overweight or obesity compared starting a GLP-1 receptor agonist against starting metformin. Incident type 2 diabetes ran 1.96% against 7.26% (HR 0.34; 95% CI, 0.26 to 0.46) and mean body-mass index change was −2.66 against −1.36 kg/m².[7] A hazard ratio of 0.34 for diabetes prevention is far larger than the 31% relative reduction metformin earned in a randomized trial, which is the usual signature of a cohort comparing people who could obtain one drug against people who could only obtain the other.
A second retrospective comparison, in 117 psychiatric outpatients over twelve weeks, found liraglutide associated with a greater percent weight reduction than metformin (estimate −3.45%; 95% CI, −5.35 to −1.55; p < 0.001). It also recorded the number that complicates the ranking: early discontinuation differed by drug (p = 0.017) and was lowest on metformin at 39.1%, against 72.2% on phentermine-topiramate.[10] Twelve weeks is too short to conclude much, but the direction matches the fifteen-year finding — the weakest drug is the one people were still taking.
What the guidelines do with metformin
The 2026 American College of Physicians living guideline on pharmacologic treatment of overweight and obesity ranks semaglutide and tirzepatide as first-line on moderate-certainty evidence, then phentermine-topiramate, liraglutide and naltrexone-bupropion. Metformin does not appear on the list at any line.[8] That absence is the position: it is a diabetes drug with a weight side effect, not a weight drug, and prescribing it as one is off-label.
The cost comparison, without the flattering framing
Metformin is off-patent and multi-source; a GLP-1 is neither. The gap that follows is enormous, and it is worth being precise about what causes it. An economic evaluation modeling manufacturing costs put sustainable cost-based prices for GLP-1 receptor agonists at $0.75 to $72.49 per month, far below market prices in every one of the 13 countries surveyed.[9] The distance between metformin’s price and semaglutide’s is not a manufacturing distance. It is exclusivity, and it is temporary, which is the subject of the patent article.
On the other side of the ledger, the cheap drug has to be taken for a long time to produce a small result, and the expensive one produces a large result for as long as it is paid for. Neither of those is a cost-per-pound calculation a landing page will do honestly, and the structure of the annual bill is set out in the annual cost article with a working version in the cost calculator.
What would actually settle it, and why it has not been run
The trial the question deserves would randomize adults with obesity to metformin or to a GLP-1 and follow both arms for a decade with weight, not glucose, as the endpoint. Nothing like it exists or is enrolling. The Diabetes Prevention Program took 3,234 participants and a national research budget to answer a narrower question about diabetes onset, and it took fifteen years of follow-up to produce the weight finding above.[1][3] No commercial sponsor has a reason to fund the comparison, because one arm is generic.
In the meantime the two are frequently prescribed together rather than against each other. Every SURPASS-EARLY participant was already taking metformin when randomized; what that trial tested was what to add to it.[6] A reader choosing between the two is often describing a choice their prescriber does not recognize.
What this comparison does not decide
Metformin has its own contraindications, its own gastrointestinal profile, and its own reasons to be stopped, all of which belong to a prescriber holding a renal function result. Its role in polycystic ovary syndrome is a separate literature covered in the PCOS article, and the two drugs are frequently taken together rather than instead of each other.
One limit applies to the whole comparison. Every GLP-1 figure above came from a trial of an approved product. Compounded semaglutide and tirzepatide are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before dispensing, so a cheaper compounded vial is not a cheaper version of the trial result. How figures on this site are established before they are published is described in the methodology.