Key takeaways
- The Premium Bonds prize rate is 4.35% tax-free from the September 2026 draw, with odds of 21,000 to 1 on every £1 Bond. NS&I says both are variable.
- That 4.35% is a mean. On the October 2026 prize distribution, the median annual win on a £10,000 holding comes out at £350, which is 3.50% of the money.
- On £1,000 of Bonds the median annual win is £0, because there's a 56.5% chance of taking nothing at all across 12 monthly draws.
- Between 65% and 69% of holders at every balance from £1,000 to £50,000 come in below the 4.35% mean over a year.
- Over 90% of UK taxpayers pay no savings tax, and inside that allowance a tax-free prize has no advantage over a taxed account paying 4.24%.
What the premium bonds prize rate pays, and what the median holder gets
The Premium Bonds prize rate is 4.35% tax-free from the September 2026 draw. The Premium Bonds odds are 21,000 to 1 for every £1 Bond in each monthly draw. NS&I's own product page states that "the prize fund rate, prize values, odds of winning and the way we allocate the number of prizes of each value are all variable, so they can change at any time."
If you want to know what that rate pays on your own money, you first have to decide which number you mean. The 4.35% is the share of all eligible Bonds handed back as prizes. It's an average across the entire holder base, not a yield on your account.
On £10,000, the mean works out at £435 a year. The median, meaning the outcome half of holders beat and half fall short of, works out at £350. That's 3.50%, or 0.85 percentage points below the headline.
Here's what the October 2026 prize distribution produces at five balances. The mean column is simply 4.35% of the holding. The median column is calculated from the actual prize counts in that draw, set out further down.
| Holding | Mean a year | Median a year | Median as a rate | Chance of no prize in 12 draws |
|---|---|---|---|---|
| £1,000 | £43.50 | £0 | 0.00% | 56.5% |
| £5,000 | £217.50 | £150 | 3.00% | 5.7% |
| £10,000 | £435 | £350 | 3.50% | 0.3% |
| £25,000 | £1,087.50 | £925 | 3.70% | effectively nil |
| £50,000 | £2,175 | £1,900 | 3.80% | effectively nil |
The chart above plots that median rate column. Find your own row and the shape is the same everywhere. The median sits below the mean at every balance, and the gap closes as the holding grows. At the £50,000 maximum the median is 3.80%, still 0.55 percentage points short of the rate on the tin. No balance reaches 4.35%, because no individual holder can.
Two jackpots a month pull the average above almost everybody
The reason is in the prize table. In the October 2026 draw NS&I paid 6,546,849 prizes worth £498,378,775. Two of those prizes were £1 million.
| Prize band | Prize value | Number in the October 2026 draw |
|---|---|---|
| Higher value, 10% of the fund | £1,000,000 | 2 |
| £100,000 | 96 | |
| £50,000 | 191 | |
| £25,000 | 382 | |
| £10,000 | 957 | |
| £5,000 | 1,914 | |
| Medium value, 10% of the fund | £1,000 | 19,934 |
| £500 | 59,802 | |
| Lower value, 80% of the fund | £100 | 2,371,145 |
| £50 | 2,371,145 | |
| £25 | 1,721,281 |
Divide the fund by the prize count and the average prize is £76.12. But 98.7% of prizes were £100 or less, and 80% of the money sits in that bottom band. The top of the table is where the mean comes from, and it's a part of the table almost nobody touches.
That's why a Premium Bonds calculator built on 4.35% overstates what most people get. The distribution is skewed, so the mean lands well above the middle. Work out the median instead and the answer drops.
The method here is straightforward. Each £1 Bond has a 1 in 21,000 chance per draw, so a £10,000 holding expects about 5.7 wins over 12 draws. Each win is a draw from the prize table above. Compound those two and you get the full distribution of a year's winnings, which is where every median in the first table comes from.
The middle half of outcomes is wider than most people expect. On £1,000 it runs from £0 to £50. On £10,000 it runs from £250 to £475. On £50,000 it runs from £1,600 to £2,275. A quarter of holders land outside those bands in each direction, and the ones above them are mostly there because of a single £1,000 prize.
One cross-check is worth doing, because it confirms the figures are internally consistent. Multiply 6,546,849 prizes by odds of 21,000 to 1 and you get about £137.5 billion of eligible Bonds. Take the £498,378,775 paid out, divide by that, and annualise: 4.35%, to four decimal places.
Where the premium bonds prize rate has been since August 2025
The increase to 4.35% was the second of 2026, and it followed a cut earlier in the year. NS&I's press releases give the whole run.
| From this draw | Prize rate | Odds per £1 Bond |
|---|---|---|
| August 2025 | 3.60% | 22,000 to 1 |
| April 2026 | 3.30% | 23,000 to 1 |
| July 2026 | 3.80% | 22,000 to 1 |
| September 2026 | 4.35% | 21,000 to 1 |
Four changes in little over a year is not a yield that holds still. Anyone comparing Premium Bonds with a fixed-term account is comparing a variable number with a contracted one. The September 2026 step added around 308,000 extra prizes and raised the pot by about £63 million, to over £497 million.
It's worth knowing who sets that number. NS&I describes itself as the UK government savings bank, and it works to a Net Financing target rather than to a market benchmark. For 2026 to 2027 that target is £15 billion, give or take £4 billion, against a stock of £252.6 billion at the end of June 2026. The first quarter delivered £1 billion. That's a different constraint from a bank repricing against Bank Rate, held at 3.75% in September 2026.
Premium Bonds versus a savings account after the April 2027 savings tax rise
Prizes are free of UK Income Tax, so the fair comparison is a gross rate that nets down to the same thing. The tax on savings interest from April 2027 rises by 2 percentage points across every band. The basic rate goes from 20% to 22%, the higher rate from 40% to 42%, and the additional rate from 45% to 47%.
Run the taxable equivalent rate both ways and the tax rise moves the bar, though not by much.
| Tax band on savings | Gross rate matching the 4.35% mean | Gross rate matching the 3.50% median on £10,000 |
|---|---|---|
| Basic, 22% from April 2027 | 5.58% | 4.49% |
| Higher, 42% from April 2027 | 7.50% | 6.03% |
| Additional, 47% from April 2027 | 8.21% | 6.60% |
Under the current 20% basic rate the matching gross figure is 5.44%, so April 2027 adds about 0.14 percentage points to it. The right-hand column is the one that bites. A basic-rate taxpayer with £10,000 in Bonds needs a taxed account at 4.49% to match the median, not 5.58%.
Against that, the Bank of England's effective rate on household deposits for August 2026 was 1.65% on sight deposits and 4.24% on new time deposits. NS&I's own Direct Saver paid 3.75% from August 2026, and its one-year Guaranteed Growth Bond 4.82%. So a holder who locks money up for a year can contract for more than the Premium Bonds median without leaving NS&I, and keeps the same HM Treasury backing.
The strongest objection: most holders get nothing from the tax exemption
The counter-argument to all of this is that the taxable-equivalent table assumes the reader pays tax on savings interest, and most don't. HM Treasury's own note on the changes says that "over 90% of UK taxpayers do not pay savings tax."
The reason is the allowances. A basic-rate taxpayer can take £1,000 of interest with no tax, a higher-rate taxpayer £500. Anyone whose non-savings income is under £17,570 gets a further £5,000 at a 0% starting rate, on top of the £12,570 Personal Allowance. Interest inside an ISA is free of tax whatever your income, and that exemption doesn't depend on the allowances at all, which is why the cash ISA limit belongs in the same comparison.
Put numbers on it. At 4.24%, a basic-rate taxpayer's £1,000 allowance covers about £23,585 of deposits. A higher-rate taxpayer's £500 covers about £11,792. Below those balances the tax-free prize carries no advantage at all, because the alternative wasn't being taxed either. The holder is swapping a contracted 4.24% for a median 3.50% and buying a lottery ticket with the difference. What that contracted rate pays next turns on whether savings rates go up in 2026, and so far the fixed-bond averages have risen while easy access has fallen.
That objection cuts the other way above the allowances, and it's why the £50,000 cap matters. A higher-rate taxpayer holding the full £50,000 is well past £11,792 of interest-generating balance, and for them the 3.80% median is genuinely tax-free where a 3.80% account wouldn't be. The exemption is worth something precisely to the holders for whom the median is closest to the mean.
There's a second objection, and it's about what is being bought. The two £1 million prizes are part of the appeal, and a holding of £50,000 carries roughly a 1 in 114,570 chance of taking one in a year. The stake itself is repayable, so what's at risk is the return rather than the capital. Measuring Premium Bonds purely as a savings rate prices that jackpot at close to nothing, which is arithmetically right and may not be how a holder sees it.
What this calculation cannot tell you
The medians here rest on one month's prize table. NS&I publishes the October 2026 distribution and an estimate for November 2026, and those two already differ: 6,546,849 prizes against an estimated 6,564,673. Prize counts move every month with the stock of Bonds, so a median is a snapshot, not a constant.
The model also treats each £1 Bond as an independent 1 in 21,000 chance in each draw. That's an approximation. NS&I sets the total prize count as the eligible £1 Bond units divided by the odds, so a fixed number of prizes is shared across a fixed pool rather than each Bond flipping its own coin. With roughly £137.5 billion of Bonds and 6.5 million prizes the distortion is far too small to move any figure in the tables, but it isn't zero.
Nothing here is a forecast. The prize rate has changed three times since the August 2025 draw, and the September 2026 figure of 4.35% was still current on NS&I's product page on 5 October 2026. It may not be by the time you read this, and NS&I states plainly that it can change the rate, the odds and the prize mix at any time. The historical sample is also short: four rate settings is not a distribution.
The taxable-equivalent columns assume a flat marginal rate on the whole of the interest. Real tax bills rarely work that cleanly, because a chunk of interest often falls inside an allowance and the rest above it. Treat those columns as the rate on the taxable slice, not on the whole balance. The comparison also ignores inflation entirely, which erodes a 4.35% mean and a 3.50% median at exactly the same pace.
One thing the arithmetic says nothing about is safety. Premium Bonds carry HM Treasury backing on the whole balance, where a bank deposit is covered to £120,000 under the FSCS protection limits. For a £50,000 holder that difference is theoretical, and the comparison tables above assume it away.
What would change the conclusion
The median moves if NS&I changes the shape of the prize table rather than its size. The 10/10/80 split between higher, medium and lower bands is a policy choice, not arithmetic. Shift money from the jackpots to the £25 and £50 prizes and the median rises towards the mean with no change in the prize rate at all. That's the lever to watch, and it's published monthly.
The gap also narrows on its own as a holding grows. The pattern in the first table is just the law of large numbers: more Bonds means more draws, and more draws means an outcome closer to the average. If NS&I ever lifted the £50,000 cap, the median at the top end would converge further on 4.35%.
And the tax comparison flips if the allowances move. The whole of the counter-argument above rests on a £1,000 and £500 Personal Savings Allowance holding still. Cut those and the 90% figure falls, the taxable-equivalent columns start applying to real people, and a tax-free 3.50% looks different against a taxed 4.24%. A rise in the allowances would do the reverse.
The number worth tracking isn't the headline rate. It's your own realised win rate over a full 12 draws against the 3.50% or 3.80% median for your balance, because that's the figure the prize rate was never quoting.