We are eight days from the first round, when 54 senators will be chosen. The Senate Model's projection became more complete today with the new "Who leads" count, and the page now shows the benches in two ways.
The model projects each candidacy's probability of election from structural variables of state politics and from polls, with the historical error of those estimates built into the confidence intervals. In practice, if a party has only two candidates, in two different states, one with an 80% chance and the other with 50%, the party is expected to elect 1.3 senators. The number is an average across the simulations, not a realistic bench, since fractions of a senator cannot be elected.
The "Probabilities" count adds all those probabilities up to show how each party's chances are spread across the 27 states. But that sum gives less weight to whoever leads close races: a candidate ahead with a 55% chance counts as 0.55 of a seat.
That is why we added the "Who leads" count, which simply adds up the two candidates ahead in each state and marks as "in play" the seats where the lead is under 3 percentage points in the model's vote projection. On the Senate map, those seats are drawn dashed.
The two counts complement each other. "Probabilities" is the more conservative: it treats smaller chances as real and gives full weight to the model's historical uncertainty. "Who leads" trusts the current picture more and projects the benches as they stand today. Where the two diverge, the gap shows where the result is still open.
On 26 September, the 2027 benches under the two counts are as follows. The first number is the sum of probabilities, with the 80% interval in brackets; the second is "Who leads", with the range from firm leads to races within 3 points in brackets.
PL — 21.7 (18 to 25); 25 (19 to 27)
União/PP — 13.2 (10 to 16); 13 (12 to 16)
PT/PCdoB/PV — 10.7 (8 to 13); 11 (10 to 12)
MDB — 7.8 (5 to 10); 8 (6 to 10)
REPUBLICANOS — 6.9 (5 to 10); 6 (6 to 7)
PSD — 5.9 (3 to 8); 6 (5 to 8)
PSB — 4.8 (3 to 6); 4 (4 to 5)
PDT — 2.1 (1 to 4); 2 (1 to 3)
PSOL/Rede — 2.0 (1 to 4); 1 (1 to 3)
PODEMOS — 1.8 (0 to 3); 2
NOVO — 1.4 (0 to 3); 2
PSDB/Cidadania — 1.2 (0 to 3); 1 (0 to 2)
There are 13 seats in play across 12 states: AL, AP, DF, ES, MA, MG, RJ, RR, RS, SC, SP (both seats) and TO.
This update also corrects the party intervals. The table was using, for each party, an approximate formula made for the blocs rather than the party's own simulations, so the intervals were wider than the model implies: in the full Senate, PL's ran from 16 to 28 seats, where the simulations give 18 to 25. Party, federation and bloc intervals now come straight from the simulations, and the table shows the average to one decimal place.
The first change to the Senate Model has an odd name: even though this is a chamber decided by plurality vote, we need to introduce a threshold clause into our model. We noticed a side effect of counting the probability of every candidacy: a party with no genuinely competitive candidate anywhere could accumulate probability simply by running candidates in more states, and qualify for a seat according to the model.
To avoid this defect in the forecast, as of today the model adopts two thresholds:
A party can only receive the point estimate of at least one Senate seat if it has at least one candidate with a 20% or higher chance of winning in their state.
A party can only receive the "leftover" seat in the point estimate if it has a remaining candidate with at least a 20% chance of winning a seat.
The logic behind this procedure is to bring the point estimate closer to the mode of the simulations and reduce noise in how the model's results are communicated. The confidence intervals are calculated from the joint simulations and are therefore unaffected by this rule, which is being adopted more as a display rule than as a change to the model's underlying logic. Either way, the model is now considered a new version (1.01).
31 Aug 2026 · projected seats: DC 0 → 1, NOVO 2 → 1 · 202 polls (+6: AP, DF, MS, PA, RN)
SE: Delegado André David (REPUBLICANOS) 17% → 18%, Rodrigo Valadares (PL) 13% → 12%
TO: Paulo Mourão (PT) 14% → 15%
Welcome to Plano Político's 2026 Senate Model. This model is a probabilistic projection of the contest for the 54 Senate seats up in this year's elections, built from public data on election results and opinion polls. To read more about the model, see the methodology. This page will publish every change to the model as it is updated with new polls, as well as any corrections that prove necessary over the course of the campaign.
As of 29 August, the most likely composition of the Senate according to the model is as follows. The first number is how many of the 54 contested seats the party is expected to win; the second is the size of its bench in 2027, adding the seats that are not in play. In brackets, the 80% interval.
PL — 11 seats (8 to 14), 21 in 2027 (18 to 24)
PT — 7 seats (4 to 9), 11 in 2027 (8 to 13)
MDB — 7 seats (4 to 9), 8 in 2027 (5 to 10)
PSD — 5 seats (2 to 6), 7 in 2027 (4 to 8)
PP — 4 seats (2 to 6), 7 in 2027 (5 to 9)
PSB — 4 seats (2 to 5), 5 in 2027 (3 to 6)
UNIÃO — 4 seats (1 to 6), 6 in 2027 (3 to 8)
REPUBLICANOS — 3 seats (1 to 6), 7 in 2027 (5 to 10)
PDT — 2 seats (0 to 4), 2 in 2027 (0 to 4)
PODEMOS — 2 seats (0 to 3), 2 in 2027 (0 to 3)
NOVO — 2 seats (0 to 3), 2 in 2027 (0 to 3)
PSOL — 1 seat (0 to 3), 1 in 2027 (0 to 3)
REDE — 1 seat (0 to 2), 1 in 2027 (0 to 2)
PSDB — 1 seat (0 to 3), 1 in 2027 (0 to 3)
AVANTE — no seats (0 to 2), losing the one it holds today
The intervals are per party and do not add up to 54 or 81: each describes the uncertainty about that one bench, not a complete scenario.