Circle Q3 2026: 3.62% and 657.6 Million in Reserve Income

The 10-Q filed by Circle Internet Group on Nov-4-2026 prints an "Average Yield" of 3.62% for the Circle Reserve Fund and $657.624 million in reserve income for the third quarter. The three figures that CleanSky had calculated on Nov-3, without publishing them, based on a filing that does not belong to Circle—3.62%, $655.3 million, and a Reserve Return Rate of 3.5%—fall within their three bands, with the yield figure matching to the basis point. The filing is the N-MFP3, the mandatory monthly report for U.S. money market funds, and the one for the Circle Reserve Fund corresponding to September was filed on Oct-7-2026, twenty-eight days before the 10-Q: the figure was calculable since then by anyone, because the fund files it under BlackRock's registration in EDGAR and therefore does not appear in Circle's list of filings. The average of the 7-day net yields declared by the Circle Reserve Fund's N-MFP3 each month has now matched the "Average Yield" printed by Circle Internet Group in its 10-Q for three consecutive quarters and with less than one basis point of difference: 3.599% calculated vs. 3.59% printed in Q1 2026, 3.571% vs. 3.57% in Q2, and 3.621% vs. 3.62% in Q3.

What Q3 2026 figures did Circle print in its 10-Q?

Circle Internet Group presented its Q3 2026 results on Wednesday, Nov-4, in an 8-K earnings release, and the 10-Q on the same day. On Nov-3, without publishing them, CleanSky had calculated three figures with their respective bands: an Average Yield for the Circle Reserve Fund of 3.62% (band of 3.60% to 3.64%), reserve income of $655.3 million (band of 645 to 670 million), and a Reserve Return Rate of 3.5%. All three came in within their bands.

Each band had pre-written which part of the method it would invalidate if breached: the yield, the identification between the two filings; the income, the model summing the two legs of the reserve; the Reserve Return Rate, the average USDC denominator. What follows is the path taken to reach each one, which is the reusable part, and the final section details where it fell short.

What is an N-MFP3 and why doesn't the Circle Reserve Fund appear in Circle's list of filings?

Form N-MFP3 is the monthly report that every money market fund registered in the United States submits to the SEC: portfolio position by position, net assets, weighted average maturity—WAM, in days—and, what matters here, a daily reading of the 7-day yield, gross and net of fees, for each business day of the month: 64 readings in the July-September 2026 quarter, 62 in the previous one, and 61 in the first.

USDC reserves live in two places. The bulk is in the Circle Reserve Fund, a government debt money market fund managed by BlackRock Advisors under Rule 2a-7—the SEC rule for money market funds—whose assets belong entirely to Circle Internet Financial, the operating subsidiary of Circle Internet Group, according to the Deloitte attestation as of Sep-30-2026. The rest is cash in banks and, since Q2 2026, Treasury bills purchased outside the fund. The Q2 10-Q weights the fund at "approximately 84%" of reserves as of Jun-30.

That fund files its N-MFP3s under the registration of BlackRock Funds, the manager's registrant entity, with number 844779 in EDGAR—the SEC's electronic archive—and not under Circle's. Circle's filing index contains, as of Nov-3-2026, 444 documents and zero filings from the N-MFP family, because the Circle Reserve Fund signs them under the BlackRock Funds registration: the two records describe the same money in two different places in EDGAR. This same filing was used in Oct-2026 to read the distribution of the fund's repo among banks, the overnight loan with government debt as collateral that makes up the largest part of its portfolio.

Why is the "Average Yield" in Circle's 10-Q the fund's net yield?

The "Average Yield" that Circle Internet Group prints in its 10-Q for the Circle Reserve Fund is the average of the 7-day net yields that the fund declares in its monthly N-MFP3s, and it has been so in all three quarters of 2026 with less than one basis point of difference: 3.599% calculated vs. 3.59% printed in the first, 3.571% vs. 3.57% in the second, and 3.621% vs. 3.62% in the third, the latter according to the 10-Q filed on Nov-4-2026 (CleanSky calculation based on the daily readings of the filing). The two checks supporting this are handled separately.

The first is one of identity, even before yield: both filings speak of the same object and state it with the same number. The Q1 2026 10-Q gives the Circle Reserve Fund a fair value of $66.467 billion as of Mar-31; the N-MFP3 for that closing declares net assets of $66.467 billion. The Q2 10-Q gives $61.917 billion as of Jun-30; the June N-MFP3, $61.917 billion. And as of Sep-30-2026, the fund total signed by Deloitte in the attestation and the net assets declared by the fund itself to the SEC differ by only $3.11 out of 62.467 billion.

The second check determines which of the two yields Circle uses. The N-MFP3 publishes the 7-day yield in both gross and net terms, and the distance between the monthly averages of both has been 8.82 to 9.00 basis points in each of the fourteen months from Aug-2025 to Sep-2026: the fund's fee. If the "Average Yield" were the gross figure, the second quarter would have had to print 3.66%; it printed 3.57%, the net figure to the basis point. Behind each average are 61 daily readings in the first quarter and 62 in the second. In the third quarter, the average gross yield was 3.711% and the net was 3.621%, and the 10-Q printed 3.62%: this is the third time in a row, and with that, the identification between the two filings ceases to be a conjecture based on two observations.

How much did the method deviate in Circle's Q1 and Q2 2026?

Getting the fund's yield right is not enough to get the reserve income right, because reserve income also includes the cash leg and because one must convert an annualized yield to the days of the quarter. The full model is a sum of two products: average fund assets multiplied by its net yield, plus average balance outside the fund multiplied by the cash yield, all prorated by the days of the quarter over 365.

Applied retrospectively to the two quarters already published, this model yields $648.5 million in Q1 2026 compared to the $652.508 printed by Circle, 0.61% below, and $663.4 million in the second compared to $667.733 actual, 0.65% below. The error is less than 1% both times and has the same sign both times: the model underestimates, due to approximating average assets with month-end closings and because cash yields slightly more than what appears in the table. This regularity is corrected by adding the average bias of the two quarters, 0.63%, which is an adjustment calibrated over two observations.

The limit of the calibration is its size, two quarters: the 2026 10-Qs also print the Average Yields for 2025—4.26% and 4.24%—but the N-MFP3 series frozen for these calculations starts in Aug-2025 and does not allow for their recalculation.

How is Circle's reserve income calculated from the N-MFP3?

The third quarter of 2026 has 92 days, from Jul-1 to Sep-30. The calculation is repeated with those operands and the documents linked at the end.

There are two assumptions. Average assets are approximated using four month-end closings—Jun-30, Jul-31, Aug-31, and Sep-30-2026—because the N-MFP3 publishes the daily yield series, but for assets, it only provides the closing figure. June is included as it is the balance with which the quarter starts; calculated only with the three closings from July to September, average assets drop to $61.834 billion and income shifts by $0.2 million. The second assumption, the 3.0% for the minor leg, carries equally little weight: ten basis points up or down change the total by $2.9 million.

The width of the band is set by the asset approximation. A one basis point error in the fund's yield is worth $1.6 million in income; a 1% error in average assets is worth $5.6 million. The four month-end closings move between 60.717 and 62.467 billion, between 1.8% below and 1.0% above their average, so the band of $645 to $670 million—a −1.6% and +2.2% over the 655.3—absorbs the case where the actual average assets fall at either extreme of that range.

The Reserve Return Rate is the third figure: reserve income divided by average USDC in circulation, annualized. With published figures, it gives 3.52% in Q1 2026 and 3.50% in Q2, and Circle prints 3.5% in both. On Nov-3, there was no primary figure for average USDC for the third quarter, so the aggregator was used: the average of the 92 daily readings from Jul-1 to Sep-30-2026 is $73.539 billion on DefiLlama, read that day (in the second quarter, that series gave 76.725 billion vs. Circle's 76.524, a 0.26% surplus). With that denominator, the $655.3 million yields 3.54%, which to one decimal place is 3.5%. The band of 3.4% to 3.6% carried that margin precisely because the denominator came from an aggregator.

What changed in the Circle Reserve Fund portfolio in September 2026?

Within the quarter, there is a movement that weighs primarily on the fourth. The fund's 7-day net yield rose from 3.62% in the Sep-1 reading to 3.84% on Sep-30-2026, while WAM went from 9 to 17 days, Treasury bills in the portfolio rose from 11.918 to 25.362 billion dollars, and repo dropped from 80.6% to 59.1% of the portfolio. The filing does not attribute a cause to this change, and neither do we. The table shows the fourteen monthly closings since Aug-2025, with the average net for each month calculated over daily readings.

The anomaly in the series is the stretch from June to August 2026, with WAM at 4, 7, and 9 days and repo above 80% of the portfolio, values that do not appear in any of the other eleven closings. The September 2026 closing, with a WAM of 17 days and repo at 59.1%, resembles that of Oct-31-2025, which had the same WAM and repo at 59.9% with an average net of 4.09% compared to September's 3.69%: composition does not set the level, which is determined by reference rates.

September is already within the Q3 average and therefore within the 3.62% that the 10-Q printed. For the fourth quarter, it is indeed decisive, and this piece makes no claims about that: the reference is the October 2026 N-MFP3. If the average of the month's 7-day net yields reaches 3.80% or more, the September jump was a regime change and the fourth-quarter yield rises with USDC flat or down; if it stays at 3.70% or less, the 3.84% on Sep-30 was a quarter-end data point.

When did Circle publish the Q3 2026 results and when did the 10-Q arrive?

Circle announced on Oct-7-2026 that it would publish Q3 results on Wednesday, Nov-4, with a webcast at 08:00 ET, and opened the shareholder question window from Oct-28 to Nov-3. That note did not name any filing nor give a time for the publication of the figures, so the 8-K earnings release and the 10-Q were inferences. Both filings are dated Nov-4-2026 in Circle's EDGAR index, with the 10-Q for the period ended Sep-30: that is 35 days from the closing, fewer than in any of the four previous observations, which ranged from 36 to 43.

The fact that both figures arrived together is what allows the check to be closed in a single day: reserve income comes from the income statement and the fund's Average Yield from the 10-Q's reserve composition table.

What is left out of the Circle Q3 2026 prediction?

Reserve income is the top line, and a large part is gone before profit: in Q3 2026, Circle recorded $411.634 million in distribution, transaction, and other costs, and closed with $226.012 million in net income from continuing operations, a figure that includes $181.8 million from the ARC token presale. That distribution depends on contracts that neither filing details, so this calculation goes as far as the income line and stops: predicting reserve income is not predicting profit.

Nor was USDC in circulation predicted, the other factor of the business, which moves in the opposite direction to yield: $76.524 billion on average in the second quarter and $73.264 billion in the third, both figures from the 10-Q. 4.3% fewer dollars in circulation with a yield five basis points higher results in less income, which is why the $657.624 million in the third quarter falls below the $667.733 in the second. The 10-Q itself breaks down the year-over-year drop in reserve income: $106.5 million less due to the 61 basis points lost in average yield, partially offset by $52.8 million more from the 8% rise in average daily USDC. Circle's sensitivity to rates measures the same lever in the other direction: 100 basis points move income by $739 million per year according to this 10-Q, compared to 737 in the previous one.

There are three points where the sources do not say the same thing. First, the reserve denominator: the Deloitte attestation gives $73,878.9 million of USDC in circulation as of Sep-30-2026—and $73,957 million in reserve assets, which is a different object—while the DefiLlama series places circulation on that same day at $74,583 million, 0.95% more, in the reading from Nov-3-2026. The attestation is the source for reserves because it is signed by an auditor; the aggregator enters only where it provides what no primary source publishes: the daily average for the quarter, and only for the minor leg. Second, Treasury bills purchased outside the fund went from $150.8 million as of Jun-30 (10-Q) to 1,502 on Sep-4 and 1,757.7 as of Sep-30 (attestation): on Nov-3, it was estimated that if they yielded one percentage point more than cash, an average of 1,000 million would add about 2.5 million to the quarter. The 10-Q put them at 3.76% vs. 3.05% for cash, a 71 basis point difference, and with that average, the contribution comes to 1.8 million: the magnitude was correct and so was the sign. It will weigh more in the fourth quarter if the balance continues to grow.

And two items that the 10-Q itself declares non-material: EURC reserves and those of tokens with denied access, 131.2 million units as of Sep-30-2026 according to the attestation. They are part of the design of a backed stablecoin and do not move the quarter's income.

Where did the Circle Q3 2026 method fail and what changes for the fourth?

The 10-Q gives the Circle Reserve Fund a fair value of $62.467 billion as of Sep-30, with cash at $9.732 billion at 3.05% and direct Treasury bills at $1.758 billion at 3.76%, and operating metrics put the quarter's average USDC circulation at 73.264 billion. With these pieces, the model's error can be dissected, which is the only thing the success doesn't tell.

The unadjusted model gave $651.2 million and fell $6.4 million short, or 0.98%. How much of that gap is caused by each assumption depends on a data point that is not public: the average balance of Treasury bills outside the fund during the quarter. The 10-Q publishes average rates by asset class, but for the balance, it only gives the closing figure, and the bills grew within the quarter: 150.8 million on Jun-30, 1,502 on Sep-4, and 1,757.7 on Sep-30. From this, two readings emerge, and both are given here because the difference between them is the limit of what the document allows us to state.

What both readings share, and therefore the only thing that holds up, is that at least half of the model's error is caused by an assumed rate—the 3.0% for the leg outside the fund, which yielded between 3.11% and 3.16%—and not by the average assets approximated with month-end closings, which is precisely what the 0.63% bias correction was intended to cover. The implicit average assets are between 62.052 and 62.190 billion dollars compared to the 61.855 used, between 0.32% and 0.54% more. For the fourth quarter, then, what is needed is not to recalibrate the bias based on a third quarter: it is to replace the assumption, because this 10-Q already publishes the yield of direct bills, which did not exist as a line item in the second quarter's 10-Q.

And a calculation that works against the correction. The Nov-3 freeze seals 656.0 million as the central figure, with an average bias of 0.73%. That same night, still before the result, the Q1 backtest did not reproduce when recalculated—648.51 and not 647.2—and with the bias corrected to 0.628%, the central figure dropped to the 655.3 published in this piece. Against the actual 657.624, the sealed and coarser figure was closer: 0.25% below vs. 0.35%. The correction improved the arithmetic and worsened the prediction, and both fall within the 645 to 670 band.

The aggregator denominator deserves its own line, because it is a repeated and measured bias: the DefiLlama quarterly average was 0.38% above the one Circle prints in its 10-Q, after being 0.26% above in the second quarter. Same sign and same order both times, which turns the margin of the Reserve Return Rate band into something that can be narrowed: changing the aggregator denominator for Circle's moves the predicted rate from 3.54% to 3.55%, and the published one is 3.56%. All three round to the 3.5% that Circle prints.

One last data point from the document that the prediction did not cover and that changes how the headline is read: total income for the quarter was $864.888 million, 17% more than a year earlier, but reserve income fell by 7.5%, 53.6 million less than in Q3 2025. The growth comes from elsewhere: $207.264 million in "other income," of which 181.8 million is the presale of the ARC token, the one for Circle's Arc chain, recognized in this quarter.

The next appointment is the fund's October N-MFP3, which by the series pattern should be filed around Nov-6-2026—the fifth business day of the month, like Oct-7, Sep-8, and Aug-6—with the 3.70% and 3.80% threshold on the month's average net yields: from there it will be clear if the fourth quarter starts with a higher yield than the third. And the fourth-quarter calculation can be closed with the October, November, and December N-MFP3s, filed in early January 2027, before Circle publishes its annual accounts in the 10-K. That time, it will indeed be possible to write it beforehand.

Sources and links: Circle Internet Group Q3 2026 10-Q, filed Nov-4-2026 (reserve composition, Average Yield, rate sensitivity) · Q3 2026 earnings note attached to the 8-K of Nov-4-2026 (reserve income, Reserve Return Rate, average USDC) · Q2 2026 10-Q · Q1 2026 10-Q · Circle Reserve Fund N-MFP3 as of Sep-30-2026 · N-MFP series of the fund registrant in EDGAR · Circle filing index in EDGAR · Circle note from Oct-7-2026 with earnings date · Deloitte attestation as of Sep-30-2026 · Circle Reserve Fund fact sheet at BlackRock · USDC daily series on DefiLlama · Circle quarterly results · CleanSky proprietary calculations on frozen extraction from Nov-3-2026.