Orforglipron vs Oral Semaglutide: One Trial Randomized 1,698 Patients and Compared Eight Side Effects at Two Dose Levels

Orforglipron is the GLP-1 tablet most asked about right now, and "is it harder on the stomach than Rybelsus?" usually gets an answer spliced from different trials — orforglipron's nausea rate from its own programme, oral semaglutide's from another, with different populations, durations and ways of asking about symptoms. That is the confound our dose-gradient piece found for dose, and the one our SURMOUNT-5 and SUSTAIN 7 pieces avoided by reading one randomized trial. This article does the same for the two tablets.

ACHIEVE-3 (NCT06045221; Rosenstock et al., The Lancet, March 2026) randomized 1,698 adults with type 2 diabetes on metformin, 1:1:1:1, to once-daily oral orforglipron 12 mg or 36 mg, or once-daily oral semaglutide 7 mg or 14 mg, for 52 weeks, at 131 centres in Argentina, China, Japan, Mexico and the United States. The abstract describes orforglipron as "a novel non-peptide (GLP-1) receptor agonist designed for daily oral administration without food or water restrictions". The trial was open-label: patients and investigators knew which drug and dose was given. Its primary endpoint was the change in HbA1c, tested for non-inferiority, and it was funded by Eli Lilly, orforglipron's maker. Its results were posted to the registry on 15 September 2026, and six of its adverse-event terms entered our corpus the next day; every one of those 24 stored rows was re-read against the registry's posted counts on 28 September 2026, and arm, term, numerator and denominator all match.

Read this as one trial, not as our pooled estimate. Our pooled clinical rate for each effect lives at the data API and averages dozens of studies with different populations, doses and durations. This article is one study's adverse-event table, useful for one reason: randomization inside it removes the population and duration confounds that make cross-trial comparisons of these two tablets unreliable. It cannot tell you how common an effect is in general, and it is not a placebo comparison.

Two dose pairs, chosen by the trial, not by us

The registry's primary outcome names the comparisons itself: orforglipron 12 mg against semaglutide 7 mg (the "low" pair) and orforglipron 36 mg against semaglutide 14 mg (the "high" pair). Those are the protocol's pairings, not a claim that the doses are equipotent. Orforglipron started at 1 mg and semaglutide at 3 mg, each stepping up every four weeks to its target dose. Per the registry, 424, 423, 426 and 425 people started and took at least one dose, and 382, 380, 388 and 401 completed. Mean age was 53.9, 825 of the 1,698 were women, and mean HbA1c was 8.29%. Argentina and Mexico together enrolled 1,074 of the 1,698, and 1,234 participants identified as Hispanic or Latino.

Adverse events were counted from baseline to the end of safety follow-up (up to 57 weeks, the registry's stated window), and the non-serious table lists a term only if it reached 5% in at least one arm. Eight of its fifteen terms are effects we track. Six of those eight are stored in our corpus; headache and decreased appetite are posted but not stored, because our collector takes whole terms from each trial up to a fixed number of rows (24), which a four-arm trial fills at six terms. We read those two directly from the registry for this article, and they do not feed our pooled estimates.

The low pair: orforglipron 12 mg against oral semaglutide 7 mg

Every cell is the registry's count divided by the same arm's at-risk denominator, read on 28 September 2026. "Acid reflux" is the registry term "Gastrooesophageal reflux disease", and "Reduced appetite" is "Decreased appetite". Gaps are orforglipron minus semaglutide, in percentage points, from the raw counts rather than the rounded percentages.

Orforglipron's arm is higher on all eight. Two example rows, verbatim from our stored evidence: "In the "12 mg Orforglipron" arm (n=424 at risk), Nausea occurred in 112 participants (26.4%)"; "In the "7 mg Semaglutide" arm (n=426 at risk), Nausea occurred in 56 participants (13.1%)."

The high pair: orforglipron 36 mg against oral semaglutide 14 mg

Orforglipron is still higher on seven of eight, but six of the eight gaps are narrower than at the low pair (abdominal pain widens from +1.9 to +2.1 points, and headache flips sign at the same size), and none is larger than 6.2 points.

What is real and what is noise

Sixteen gaps invite sixteen stories. Running a two-proportion z-test on each row (the standard large-sample approximation, no continuity correction) and applying a Bonferroni correction for testing sixteen comparisons at once (critical |z| ≈ 2.96 for a family-wise 5%, against the usual single-test 1.96):

Nausea (z ≈ 4.86), vomiting (z ≈ 4.09) and diarrhoea (z ≈ 3.29) at the low pair: all three clear the corrected bar. Orforglipron 12 mg reported roughly twice the nausea and vomiting of semaglutide 7 mg, and that is not sampling noise.

Vomiting at the high pair (z ≈ 2.64), constipation at the low pair (z ≈ 2.60), nausea at the high pair (z ≈ 2.12) and acid reflux at the low pair (z ≈ 2.06): clear the single-test line and not the corrected one. With sixteen comparisons and no correction, the chance of at least one false "finding" is about 56% even if nothing genuinely differed. Read them as suggestive.

Everything else: within sampling noise at these sample sizes, including high-pair diarrhoea (z ≈ 1.90) and constipation (z ≈ 1.89), which sit just under the single-test line.

We publish the z-values rather than a "significant" flag so the two tables do not read as sixteen findings when they support three.

The gap closes because semaglutide's dose step moves, not orforglipron's

Reading the four arms as two dose ladders explains why the high pair separates less. Moving orforglipron from 12 to 36 mg changed no effect by more than 1.7 points (nausea 26.4% to 27.4%, diarrhoea 21.7% to 23.4%, vomiting 17.0% to 16.5%). Moving semaglutide from 7 to 14 mg raised nausea by 8.0 points (13.1% to 21.2%), diarrhoea by 5.0 (13.1% to 18.1%) and vomiting by 2.6 (7.7% to 10.4%). Those within-drug differences are descriptive — they are not part of the sixteen-test family above — but the direction is plain: orforglipron's gastrointestinal rates were already at their level at 12 mg, and semaglutide 7 mg was the gentlest arm on every effect but headache. It is the mirror of SUSTAIN 7, where the dose step of the comparator drug, dulaglutide, closed the gap.

Beyond the eight tracked effects

The abstract's own summary is that gastrointestinal events as a class occurred in 249 (59%) of 424 people on orforglipron 12 mg, 245 (58%) of 423 on 36 mg, 157 (37%) of 426 on semaglutide 7 mg and 193 (45%) of 425 on 14 mg, "most of which were mild to moderate in severity". Three other gastrointestinal terms in the registry's table follow the same pattern: dyspepsia (48, 44, 18 and 33 people in the four arms, in that order), eructation (42, 32, 17, 14) and abdominal distension (24, 34, 11, 14).

Stopping the drug and leaving the trial are different counts. The abstract reports that 37 (9%) and 41 (10%) people on orforglipron 12 and 36 mg stopped study treatment because of adverse events, against 19 (4%) and 21 (5%) on semaglutide 7 and 14 mg. The registry's participant flow counts people who left the study altogether with "Adverse Event" as the reason: 7, 4, 4 and 5. Someone who stopped the tablet but kept attending visits appears in the first count and not the second. The authors' interpretation names three things as higher with orforglipron: "the incidence of gastrointestinal events, discontinuations due to adverse events, and mean increase in pulse rate" (3.7 and 4.7 beats per minute on orforglipron, 1.0 and 1.5 on semaglutide).

Serious adverse events, from the registry: 21 of 424 (5.0%) on orforglipron 12 mg, 40 of 423 (9.5%) on 36 mg, 19 of 426 (4.5%) on semaglutide 7 mg and 24 of 425 (5.6%) on 14 mg. The high-pair gap of 3.8 points has a z of about 2.10 — descriptive, outside the tested family, and summed across 106 different serious terms, most affecting a single person. The pancreas and gallbladder terms among them are single events spread across arms: acute pancreatitis once on orforglipron 12 mg and once on semaglutide 14 mg; cholelithiasis once on each of the same two arms; cholecystitis once on semaglutide 7 mg; chronic cholecystitis once on orforglipron 36 mg. There were four deaths: one on each orforglipron dose and two on semaglutide 7 mg.

Two non-gastrointestinal rows are worth reading correctly. Hyperglycaemia was listed for 69 people (16.2%) on semaglutide 7 mg against 20 to 34 on the other arms; that is also the arm whose HbA1c fell least, and in a diabetes trial a high-blood-sugar event measures glucose control rather than a separate side effect of the drug. Diabetic retinopathy was listed for 54 to 60 people in every arm (12.7% to 14.2%), with no gradient between drugs.

On the efficacy side, which was the trial's purpose, the abstract's primary treatment-regimen estimates of HbA1c change at week 52 were −1.71 and −1.91 percentage points on orforglipron 12 and 36 mg against −1.23 and −1.47 on semaglutide 7 and 14 mg, from a baseline of 8.3%, and both orforglipron doses were superior to both semaglutide doses (estimated differences −0.48 and −0.44 points at the two protocol pairs, −0.24 for orforglipron 12 mg against semaglutide 14 mg and −0.68 for 36 mg against 7 mg; the superiority tests were prespecified as a hierarchy after non-inferiority was met). The registry's posted least-squares means use a different estimand (data up to the point a participant stopped the drug or added another glucose-lowering medicine), which is why its HbA1c figures differ (−1.91, −2.16, −1.11, −1.45); its body-weight changes on the same basis were −6.7% and −9.2% against −3.7% and −5.3%. So at both pairs, the arm with more gastrointestinal effects also lost more weight — about 3 points more at the low pair and 4 at the high pair, on the registry's figures.

What this does not tell you

No placebo arm. ACHIEVE-3 is drug versus drug, so nothing here separates either drug's effect from a background symptom rate, unlike our constipation or headache pieces.

A diabetes trial at diabetes doses. The population had type 2 diabetes on metformin, with a mean HbA1c of 8.3%. It says nothing directly about orforglipron in obesity without diabetes, or about injectable semaglutide.

One trial, open-label, sponsor-run. Neither patients nor investigators were blinded, which matters most for symptom-type events like nausea: knowing you are on the newer drug can change what you notice and report. The sponsor makes one of the two drugs. None of that changes a registry count, but all of it bounds what a count can mean.

A 5% listing threshold. Fatigue, dizziness, hair loss and the other tracked effects absent from the non-serious table did not reach 5% in any arm; that is not the same as not occurring.

Whole course, 52 weeks. Each arm includes its titration weeks, so these are at-least-once shares over the whole period, not rates at a steady dose.

The pairs are the protocol's. 12 against 7 mg and 36 against 14 mg are how the trial was built; they are not matched for potency.

Multiple comparisons, restated. Read low-pair nausea, vomiting and diarrhoea as established, the four gaps between 1.96 and 2.96 as suggestive, and the rest as noise.

If you are on a GLP-1 now

This is educational content, not medical advice, and it does not tell you which drug is right for you — that depends on your own history, your prescriber's judgement, and factors a single trial's adverse-event table cannot capture. Nothing here should be used to start, stop or switch a medication without your prescriber.

Every figure in this article was read from ACHIEVE-3's ClinicalTrials.gov results record and from the PubMed abstract of its primary publication (PMID 41765029) on 28 September 2026. The six stored effects count toward our pooled estimates for those effects as registry rows with an unspecified dose tier, so they sit in no dose-tier pool; the pooled figures at the data API move with each collection run, so verify there before citing them.