Le Millénaire · election on 3 November 2026 · 200,000 simulations per chamber · updated 3 November 2026
Chances of control on 3 November
435
House of Representatives
79.1%
chance Democrats win a majority (218 seats)
D 228.7 · R 206.3
expected seats out of 435 · 90% of cases: 207 to 252 D seats
100
Senate
48.3%
chance Democrats and allied independents win a majority (51 seats)
D+I 50.4 · R 49.6
expected seats out of 100 · 90% of cases: 47 to 54 D + I seats
independents with no Democratic opponent (Achilles, Bodnar, Osborn, Bengs) counted with the Democrats
36
Governors
D 24·R 24
expected governors out of 48 seats computed (D 23.7 · R 24.3)
D 21 to 26
90% range of Democratic governors · 36 seats up · Alaska and Maine not computed
In short
House. Democrats favoured: Democrats win the 218-seat majority in 79.1% of the 200,000 simulations, with 229 seats on average (they have held 215 since 2024).
Senate. Toss-up: Democrats and allied independents reach 51 seats in 48.3% of the simulations (50.4 seats on average, 47 today). At 50-50 the Republican vice president breaks ties.
Governors. Of the 48 seats computed, 23.7 Democrats and 24.3 Republicans expected: Republicans would keep most governorships.
Probabilities are conditional on a Democrat versus Republican contest in each race. Updated 3 November 2026.
How the election can play out
The four possible outcomes for Congress; they add up to 100%. House and Senate are simulated together, with the same national polling error: if the polls miss in one chamber, they miss in the same direction in the other.
Democrats control both chambersHouse and Senate, allied independents included45.9%
How to read. Each figure is the share of the 200,000 simulations in which the outcome happens. Assumption: the national polling error is the same for both chambers; regional, state and district errors stay independent. This link cannot be estimated from past midterms alone. The four figures are adjusted to match exactly each chamber's chances shown above. The Senate counts independents with no Democratic opponent with the Democrats.
How the chances have moved
One point per day since 1 August 2026. Points up to and including 22 October 2026 were recomputed afterwards with the current method and only the data published by each date (polls whose fieldwork had closed, finance and indicators already released). Hover or tap the chart to read each day.
House, D chancesSenate, D + allied I chances
Democrats control both chambers: 45.9% (joint simulation of both chambers, see “How the election can play out”).
The electoral map
The three elections state by state. Hover or tap a state for its figures.
House · 435 seats
Colour: share of the state's seats expected for each side (sum of each district's chances).
Categories: Toss-up below a 60% chance for the favourite, Lean from 60 to 75%, Likely from 75 to 95%, Safe above. Independents with no Democratic opponent, Achilles (Idaho), Bodnar (Montana), Osborn (Nebraska) and Bengs (South Dakota), are shaded like the Democrats, with whom they are counted. Alaska and Maine (governors): multi-candidate contests, not yet probabilized.
The House of Representatives
Distribution of Democratic seats over 200,000 simulations. Blue: Democratic majority (218 seats or more).
National generic ballot (likely voters): each dot is a poll published since June, as the Democratic minus Republican margin. The line is a descriptive 21-day rolling average. The model applies its own aggregation rule.
74 polls from 30 pollsters, 17 June 2026 to 30 October 2026. Value used by the model on 3 November 2026: Democrats +7.3 points over Republicans, before the polling bias correction. Sources: Decision Desk HQ until 1 October, then Wikipedia and PollsMax, collected daily.
The test on past elections
Each engine was replayed on past midterms, with only the data available at the test date.
Engine
Test
Error
Probabilities
House, national
2010, 2014, 2018 at day −60
1.46 points of Democratic share
error carried into the 200,000 simulations
House, districts
984 contests 2010-2018, out of sample
margin: RMSE 9.86 points (10.18 without finance)
Brier 0.062
House, polled districts
214 districts 2014-2022, leave one year out
margin: RMSE 7.30 points with polls (12.05 without)
correct winner 80.4% (72.9% without)
Senate
132 races 2014-2022, leave one year out
margin: mean error 4.58 points, RMSE 5.92
correct winner 90.9%
Governors
144 races 2010-2022, leave one year out, from 45 days before the vote to election day
margin: mean error 6.66 points at day −45, 5.77 at day −30, 5.31 at day −21, 3.94 at day −7, 3.71 on election day
correct winner 86.8% at day −45, 91.0% at day −30, 93.8% on election day · Brier 0.086 → 0.052
Senate, incumbents and former officeholders
148 races 2014-2022, leave one year out, 17 days before the vote
polled races 2014-2022, leave one year out, 17 days before the vote
mean polling error: Senate 5.01 → 4.49, governors 5.17 → 4.54, House 7.27 → 5.69 points
Senate correct winner 91.0% → 93.1%
The generic ballot against the results
Midterm
Polls at day −45
Actual national vote
Gap, vote − polls
2006
D +9.2
D +8.0
R +1.2
2010
R +3.6
R +6.8
R +3.2
2014
R +1.8
R +5.7
R +3.9
2018
D +8.8
D +8.6
R +0.2
2022
D +1.8
R +2.7
R +4.5
Democratic minus Republican margin, in points. Polls overstated Democrats in 5 of 5 midterms; the model therefore takes off every poll the average 2006 to 2022 midterm gap at the same distance from election day (2.95 points a month out, 2.26 on the eve). Sources: FiveThirtyEight average, Clerk of the House.
The 2024 election
The starting point: the Congress and governors elected in 2024.
House
D 215 · R 220
Senate
D 45 · I 2 · R 53
Governors
D 23 · R 27
220
Republican House seats
215
Democratic seats
+3
seats Democrats need to gain
+4
D + I seats to gain in the Senate
House: national vote 49.8% Republican, 47.2% Democratic. Senate after 2024: 53 Republicans, 45 Democrats, 2 independents who caucus with them. Sources: Wikipedia, from the Clerk of the House and official state returns.
How it works
A forecasting model is not one more poll. It gathers what can be measured: published polls, past results, incumbency, fundraising. From these it derives a distribution of possible outcomes, with their probability.
Three engines
House, Senate and governors each have their own engine, validated on past midterms. All start from the same national environment, the likely-voter generic ballot, recomputed daily.
Past polling errors
Polls missed in every midterm, most often in favour of Democrats. The model therefore takes off the polls their average overstatement of Democrats in the 2006 to 2022 midterms at the same distance from election day (2.26 to 2.95 points), then replays the size of the remaining errors in both directions.
Two hundred thousand simulations
Each chamber is replayed 200,000 times, with a national error shared by every race and an error specific to each race. The probability shown is the share of simulations in which the event happens.
What remains outside the calculation
House and Senate are simulated with the same national polling error, an assumption that cannot be checked over four midterms. Candidates outside the top two are bounded by the polls, and Alaska and Maine (governors) are not yet probabilized.
What the model measuresThree independent engines, validated out of sample on past midterms. House: each district's 2024 result, incumbency and the national generic ballot, with a national error calibrated on polling history. Senate: the method of all Millénaire forecasts, in three blocks. The first is the polls: one per pollster, weighted by age, sample size and pollster accuracy, and corrected when commissioned by a party, a candidate or an allied group. The second is state history, with the edge of an incumbent or a former officeholder. The third is structure. The weight of the polls rises as the vote approaches. Governors: structural formula (2024 presidential vote, national climate and an incumbent governor on the ballot) blended with each race's polls at every point of the campaign; the polls' weight rises from about 68% two months out to 88% on election day, learned on four elections.
What it does not say yetChances are computed between Democrat and Republican, or between independent and Republican where the independent is the main challenger (Idaho, Montana, Nebraska, South Dakota). 14 districts (Louisiana, same-party runoffs) remain outside the full field. In the Senate, a third candidate's chance is bounded by the polls: below 0.5% in Alaska, Maine and Nebraska; at most: Montana 2.2% (Alani Bankhead, D). The joint House and Senate probability assumes a national polling error shared by both chambers: this link cannot be measured from available history.
Sources
Polls: Decision Desk HQ until 1 October, then Wikipedia, 270toWin and PollsMax, collected daily, with pollsters’ own releases; pollster reliability ratings from the FiveThirtyEight public archive; 265 of the 373 polls used since 1 September checked against the pollster’s or sponsor’s own release. Past results: Clerk of the House, Federal Election Commission, MIT Election Data and Science Lab. Fundraising: Federal Election Commission public files. Candidates: state election authorities. Generic ballot history: FiveThirtyEight.
Credits
All rights belong to Le Millénaire SAS, 47 rue de Vivienne, 75002 Paris.
Project director: William Thay Deputy project director: Sean Scull / Pierre Clairé Political and data analysis: Matthieu Hocque / Augustin Bataille Data analysis, mathematics and validation: Florian Gérard-Mercier