The federal debt's 3.447% average hides 13 rates from 1.127% to 7.577% — and Treasury now pays more on bills than bonds

You have almost certainly seen the one number: the U.S. government pays an average of 3.447% on its interest-bearing debt as of July 31, 2026 — a figure widely reported as a multi-year high. That one number is an average of 13, and we read all 13. They do not cluster around it, and the order in which the government pays the three most familiar of them has flipped since 2020 — neither of which the single figure can show you. Here is the ledger that number summarizes. Treasury's Bureau of the Fiscal Service publishes, free and without a key, the monthly dataset "Average Interest Rates on U.S. Treasury Securities" — 307 monthly records back to January 2001, latest record dated 2026-07-31 (retrieved 2026-08-15). For each month it lists, for every category of interest-bearing federal debt outstanding, the average rate the government is currently paying on the stock — the coupon load carried across every vintage still outstanding, not the yield to issue new debt. For July 2026 the dataset lists 13 instrument categories plus three subtotals. The blended figure — "Total Interest-bearing Debt" — is 3.447%. Its two halves: Total Marketable 3.443%, Total Non-marketable 3.463%. The 13 underlying categories and their average rates: Treasury Bills 3.758%, Treasury Notes 3.309%, Treasury Bonds 3.442%, Treasury Inflation-Protected Securities (TIPS) 1.127%, Treasury Floating Rate Notes (FRN) 3.948%, Federal Financing Bank 2.383%; and, among the nonmarketable series, Domestic Series 7.577%, Special Purpose Vehicle 2.956%, State and Local Government Series 3.353%, United States Savings Securities 3.152%, United States Savings Inflation Securities 4.466%, Government Account Series 3.461%, Government Account Series Inflation Securities 1.392%. Across those 13 categories the average rate runs from 1.127% (TIPS) to 7.577% (Domestic Series) — a spread of 6.450 percentage points, around a blend of 3.447%. The highest rate is 6.72 times the lowest: 7.577% divided by 1.127% (a ratio of the two rates, not of the blend). The lowest line sits 2.32 points below the blend; the highest sits 4.13 points above it. The sharpest structure is inside the marketable book — the tradable debt Treasury groups under its Total Marketable line. Ranked by average rate, its six categories run FRN 3.948% (highest), Bills 3.758%, Bonds 3.442%, Notes 3.309%, Federal Financing Bank 2.383%, TIPS 1.127% (lowest). Reading only the three plain instruments a saver would recognize — bills, notes, bonds — the order is inverted from the textbook: Treasury Bills (maturities of a year or less) average 3.758%, higher than Treasury Bonds (20–30 years) at 3.442% and Treasury Notes (2–10 years) at 3.309%. Bills exceed notes by 0.449 points and bonds by 0.316 points. That ordering is a reversal, and it is datable. In January 2020 the same three rose in maturity order — Bills 1.683%, Notes 2.144%, Bonds 3.841% — bonds costing 2.158 points more than bills. The pattern held in January 2021 (0.109% / 1.713% / 3.364%) and January 2022 (0.103% / 1.396% / 3.016%). Then it flipped: in January 2023 (Bills 4.242%, Notes 1.753%, Bonds 3.022%), January 2024 (5.411% / 2.340% / 3.106%), January 2025 (4.455% / 2.891% / 3.236%), January 2026 (3.760% / 3.169% / 3.369%), and again in the latest July 2026 ledger, Bills carried a higher average rate than both notes and bonds every time. The floating-rate line moved most of all: FRN averaged 0.234% in January 2022 and 3.948% in July 2026, a 3.714-point rise.

The following is analysis, not fact. A blended debt rate is a weighted average, and an average conceals its own dispersion by construction. We read all 13 lines the single number stands in for. The reported 3.447% is doing the work of 13, and the 13 do not cluster around it: they span from a TIPS book still averaging 1.127% to a Domestic Series line at 7.577%. Inside the marketable book the six categories spread 2.821 points end to end — FRN 3.948% down to TIPS 1.127% — and even the three plain instruments a saver would recognize sit out of maturity order, bills 3.758% above bonds 3.442% above notes 3.309%. You see one rate on the debt. We read the 13, and the inversion only appears when you do; the blend cannot show it. Why the tradable curve is inverted by cost is a composition story, and here it is a hypothesis ^mmodeled, not an observed fact. The one thing the dataset does establish is what it measures: the average rate on the stock outstanding, not the yield on new issuance — so each category's number is the coupon load of every security in it that has not yet matured. Two mechanics follow from Treasury's own definitions: bills mature in a year or less and Floating Rate Notes reset to the current short rate, so both reprice to prevailing short rates quickly, while notes (2–10 years) and bonds (20–30 years) turn over slowly and keep older coupons on the books for years. A stock that reprices fast will track recent rates; a stock that reprices slowly will lag them. That is enough to make the observed ordering — short, fast-repricing debt averaging above long, slow-repricing debt — consistent with short rates having risen since 2020 and long-dated coupons issued before then still outstanding. What we do NOT do is measure the dollar volume or coupon of any specific issuance vintage; this run pulled no balances, so the size of the cheap long-dated stock is asserted here as the likely driver, not shown. It is a claim about the composition of the existing stock, not about today's market yield curve. The multi-year anchors show the inversion is neither a one-month artifact nor brand-new: it appears in every January from 2023 through 2026 and in July 2026, after being absent in 2020, 2021 and 2022. Across those same anchors the gap has also compressed at the ends we sampled. We read the January triple and the July record, not every month between, so we scope to the anchors: the bills-minus-notes gap was 3.071 points at its January-2024 anchor peak (bills 5.411%, notes 2.340%) and 0.449 points in July 2026 (bills 3.758%, notes 3.309%) — bills easing while notes climbed. We do not assert the fall was continuous month to month; we did not pull the intervening series. What the whole ledger shows that the headline cannot is that "the interest rate on the federal debt" is a convenient fiction: there are 13 of them, they range across 6.45 points, and the order in which the government pays the three plain instruments has reversed since 2020.

The load-bearing caveat is what the number is. Average interest rate here means the rate on the debt already outstanding, weighted across vintages — a backward-looking cost measure, not the yield Treasury would pay to borrow today. The marketable inversion (bills above bonds) is therefore a statement about the composition of the existing stock, and it must not be read as "the Treasury yield curve is inverted," which is a separate, market-priced claim this dataset does not make. The two can point in different directions. Second, this dataset carries no dollar balances, so every claim here is about rate, not weight. The 7.577% Domestic Series and the 4.466% Savings Inflation Securities are the highest-rate lines, but they are small, largely legacy nonmarketable series; a high average rate on a tiny balance moves the blend very little, and we do not assert otherwise. To rank instruments by total dollars of interest paid would require joining the Monthly Statement of the Public Debt's outstanding balances, which we did not pull this run — so we scope strictly to the average-rate figures the table publishes. Third, the category set is not fixed. Treasury's own list changes over time: a "Foreign Series" line present in January 2022 has run off by July 2026, and a "Special Purpose Vehicle" line has appeared; the "13 categories" here are exactly those the dataset reports for the July 2026 record, no more. Rates are shown to the dataset's three published decimals, and the whole picture is a single month-end snapshot in a monthly series; a re-pull will move the third digit and, in time, the ranking. The durable claims are the structural ones — the 6.45-point spread, and bills averaging above notes and bonds in every January from 2023 through 2026 — not any single decimal.

From the vantage of the yield curve as it is usually taught, the ranking should be settled: money lent for 30 years costs the borrower more than money lent for four months, because the lender demands more to wait longer. Sorted that way, the federal debt looks orderly — and until 2023 it was: bills cheapest, bonds dearest, notes between, three Januaries running. The same table read today disagrees with the textbook, because it is measuring a different thing. It records not what the government would pay to borrow now but what it is still paying on everything it borrowed before — and on that measure the cheap decade sits at the long end, locked in, while the expensive years sit at the short end, repriced. The curve did not so much invert as get overtaken by its own history. Both tables are built from the same securities; they disagree because one asks what the next dollar costs and the other asks what the last trillion still does.