Where these numbers come from
Every IRMAA site publishes a table. Almost none says where the table came from. Ours comes from CMS and SSA directly — and the most useful thing we can tell you is the shortcut we deliberately did not take.
Why we don’t calculate the Part B surcharge
Medicare Part B has a published structure. Standard enrollees pay 25% of the programme’s true cost; enrollees subject to IRMAA pay 35%, 50%, 65%, 80% and 85% across the five tiers (POMS HI 01101.031).
That makes it tempting to compute the whole column from one number. Take the standard premium of $202.90, divide by 25% to get the true cost, apply each tier’s percentage, subtract. It is clean, it is nearly right, and it is wrong by a few cents — because the CMS standard premium includes a repayment amount under §1839(a)(6) that the shortcut ignores.
| Tier | Shortcut gives | CMS publishes | Error |
|---|---|---|---|
| Tier 1 (35%) | $81.16 | $81.20 | -0.04 |
| Tier 2 (50%) | $202.90 | $202.90 | — |
| Tier 3 (65%) | $324.64 | $324.60 | +0.04 |
| Tier 4 (80%) | $446.38 | $446.30 | +0.08 |
| Tier 5 (85%) | $486.96 | $487.00 | -0.04 |
Four to eight cents is not much money. It is, however, the difference between a figure that matches the government’s and one that doesn’t — and a reader who checks us against CMS and finds a mismatch has no way to know it was only rounding. So the tables on this site carry the CMS-published amount, and the arithmetic above exists only to show you the trap.
What comes from where
| Figure | Source | Published |
|---|---|---|
| 2026 premiums, brackets and surcharges | CMS fact sheet | 2025-11-14 |
| 2025 premiums, brackets and surcharges | CMS fact sheet | 2024-11-08 |
| MAGI threshold tables | SSA POMS HI 01101.020 | Standing |
| Part B cost-share percentages | SSA POMS HI 01101.031 | Standing |
| Threshold indexing rule | Social Security Act §1839(i)(5), 42 U.S.C. §1395r(i)(5) | Statute |
We do not cite other blogs, brokers or comparison sites — only the documents they are all copying from.
How the thresholds move each year
The MAGI lines are indexed under Social Security Act §1839(i)(5), 42 U.S.C. §1395r(i)(5). The measure is the Consumer Price Index for all urban consumers (CPI-U), United States city average, over a 12-month period ending with august of the preceding calendar year, versus the 12-month period ending with august 2018 (for years beginning with 2020). the $500,000 / $750,000 top-tier floors are not indexed until calendar years beginning after 2027, using august-window cpi-u vs august 2026.
The result is then rounded: if any dollar amount after being increased is not a multiple of $1,000, such dollar amount shall be rounded to the nearest multiple of $1,000.
This is why a projection is possible before CMS publishes — and why any projection that doesn’t state its CPI window is a guess dressed up as arithmetic.
One important limit: the $500,000 and $750,000 top-tier floors are not indexed until years beginning after 2027. Rolling every line forward by CPI produces a top tier that is simply wrong. Any projection we publish will hold those floors fixed and label every projected figure as projected.
Reading the brackets correctly
CMS does not use one interval rule for every row, and the difference matters if your income lands exactly on a line:
| Rows | Rule |
|---|---|
| Standard tier | MAGI at or below the limit |
| Tiers 1–3 | Above the lower figure, at or below the upper |
| Tier 4 | Above the lower figure and strictly below the top floor |
| Top tier | MAGI at or above the floor |
So $109,000.00 exactly is still the standard tier — one dollar more is not. Our table applies these rules per row rather than assuming a single convention.
The marriedSeparate column applies only to those married filing separately who lived with their spouse at any time during the tax year. They skip tiers 1-3 and go straight from standard to the 80% and 85% tiers. Those who lived apart the entire year use the single table (POMS HI 01101.020; SSA 1839(i)(3)(C)(iii)).
Corrections
If a figure here disagrees with CMS or SSA, the source wins and we are wrong. Every number traces to a named document above, which makes any error checkable rather than a matter of opinion — and fixable in one file.