Federal MRP, May–July 2026

01

The first model in which nobody's range reaches a majority

Seventy-six seats carry the House of Representatives. The highest number any party reaches at the top of its own published range in this model is 72, and the two models before it both put Labor above 76 across their whole bands.

Each bar is the published seat range, drawn end to end. The line marks the 76 seats a majority needs.

Firm, lean and in doubt sum to the total leading for every party, and the totals sum to 150. In doubt is a probability of victory below 70 per cent, lean 70 to 95 and firm above 95. The seat ranges come from a Monte Carlo analysis of 20,000 simulations and the report describes them as a 95 per cent confidence interval.

The three models trace a straight line on seats and a curve on votes. Labor goes 98, 83 and 68, losing fifteen seats and then fifteen more. One Nation goes 12, 52 and 60, which is a leap and then a step. The Coalition goes 29, 9 and 13, falling by twenty and recovering by four. On this third reading the two largest parties are eight seats apart with 128 of the 150 between them, and the party that held 43 seats fourteen months earlier has thirteen.

A hung parliament is not an incidental feature of this model, it is the whole distribution. Labor's band of 60 to 72 never touches 76 and One Nation's of 53 to 66 never comes within ten of it. Adding the Greens' three and the Others' six to Labor's 68 reaches 77, one seat over, which is the arithmetic that makes the crossbench the subject rather than a footnote. On the central forecast the parliament's balance rests on the nine members the model places outside the three largest parties, and six of those nine seats it classifies as lean or in doubt.

The classifications say the same thing about firmness as the previous model did. Labor's 68 is 40 firm, 18 lean and 10 in doubt; One Nation's 60 is 27 firm, 25 lean and 8 in doubt. Fewer than half of One Nation's seats are firm, against nearly three-fifths of Labor's, and the Coalition's thirteen are only four firm. These are modelled probabilities rather than measured ones, and the report publishes no interval on the vote shares they rest on beyond the ±5 points it states for each electorate.

02

One Nation passes Labor on the modelled primary vote

Across the three models One Nation goes 17 per cent, 27 and 30 while Labor goes 33, 29 and 26. This is the first of the three in which the modelled One Nation share is the largest in the country.

The bars show first preferences at the May 2025 election and in the three MRP models, in per cent.

Against the May 2025 election this model puts One Nation up 23.6 points, the Coalition down 10.8, Labor down 8.6, the Others down 5.0 and the Greens up 0.8. The report publishes no national two-party preferred figure for this wave or the one before it; only the first of the three models published one.

The shape of the three readings matters more than the latest number. One Nation's rise is decelerating: ten and a half points between the first two models and three between the second and third. The Coalition's fall has stopped at 21 per cent across two models. The Greens have moved between 12 and 13 throughout and the Others have fallen steadily from 15.0 to 10. The only series still travelling in a straight line is Labor's, down four points and then three.

Four points now separate One Nation from Labor, where the second model had Labor two ahead. That crossover is the headline, and it should be read with the same caution the whole page carries: these are modelled shares and the report publishes no interval against them. What is firmer is the gap at the other end. One Nation is nine points clear of the Coalition in this model, where in the first it was seven points behind, which is a sixteen-point reversal of the right-of-centre order inside eight months.

Retention explains the direction. Of the people who say they voted Coalition in May 2025, 56 per cent would vote Coalition again and 39 per cent would vote One Nation, against 56 and 37 in the previous model — so the Coalition's leakage has stabilised rather than stopped. Labor's retention falls from 76 per cent to 69 and the share of its 2025 voters choosing One Nation rises from 13 to 17. One Nation holds 97 per cent of its own in both models, and the Greens 82 then 84.

03

One Nation rises in all twenty-six groups the model reports

This report tabulates all three models side by side for every cohort it publishes. Across gender, age, education, income, location and jurisdiction, not one group moves against the trend.

Every full cohort row in the report sums to 100 across the five parties. The three columns are the report's own figures and the change column is the difference between the first and the third. These are modelled estimates and the report publishes no interval for them.

Rural and regional Australia moves furthest, from 24 per cent to 41. The inner suburbs move least in points, from 10 to 19, tied with the highest income band at nine; but a rise from 10 to 19 is a near-doubling, and it matters because the first model had One Nation in single figures there. The widest gap in the table is between those two groups at the latest reading: 41 against 19, which is twenty-two points and the structural fact that the seat projection in section 01 is built on.

Education continues to divide the vote more sharply than income. The spread across the five education bands is 19 points, from 39 among those who did not finish Year 12 to 20 among postgraduates. Across the five income bands it is four points, from 31 to 27, and the highest income band is no longer the outlier it was: it sat at 19 per cent in the second model and 27 in this one, the largest single-wave rise of any group in the table.

The jurisdictions have converged on each other. In the first model One Nation ran from 6 per cent in the Australian Capital Territory to 20 in Queensland, a fourteen-point spread. Here it runs from 16 to 34, which is eighteen points, but seven of the eight jurisdictions now sit between 27 and 34. A vote that was concentrated in Queensland and the Northern Territory in October 2025 is a national vote with a Queensland premium by July 2026.

04

The Coalition's regional advantage has gone, and its income advantage has not

In the first model the Coalition polled best outside the capitals. In this one it polls the same everywhere, at 21, 21 and 20 per cent across the three location bands, while its figure among the highest earners stays eight points above its national share.

The Coalition's lowest figure anywhere in this model is Tasmania at 13 per cent and its highest is 29 among those earning more than $200,000. Its state figures are New South Wales 20, Victoria 24, Queensland 21, South Australia 19, Western Australia 19, Tasmania 13, the Australian Capital Territory 16 and the Northern Territory 18.

The three location rows are the ones to read first. A party that led the field in rural and regional Australia at 26 per cent in October 2025 now polls 20 there, below its figure in both metropolitan bands and twenty-one points behind One Nation. The flatness is not a recovery in the cities; it is the bush falling to meet them. Across the whole model there is now no place where the Coalition's vote is meaningfully above its national 21 except at the top of the income distribution.

That income row has its own shape. The Coalition's figure among those earning more than $200,000 went 32, 34 and 29 across the three models, rising while everything else fell and then falling eight points in the latest reading. One Nation's figure in the same band went 18, 19 and 27. Whatever insulated the Coalition's highest-income support through the first two models stopped insulating it in the third, and that is the single largest One Nation gain in any group between the last two models.

Education is where what remains of the Coalition's vote now sits. It takes 25 per cent of undergraduates and 23 of postgraduates against 18 of the TAFE-qualified and 19 of those who did not finish Year 12. That ordering is the reverse of the one the party held for most of the last half-century, and it arrives not because degree holders moved towards the Coalition — their figures fell too — but because the groups below them left faster.

05

The preference model keeps adding contests it did not need before

The first model borrowed two electorates' flows for every One Nation final count in the country. The second measured flows for two contest types. This one measures four, because One Nation now reaches the final count against the Greens and against independents as well.

The progression is a record of a party arriving. In October and November 2025 the model used Maranoa for Coalition-against-One Nation counts and Hunter for Labor-against-One Nation counts, and said in its own limitations that two electorates standing in for a category added uncertainty. By February and March it was using respondent-allocated flows for Labor-against-One Nation and Coalition-against-One Nation contests and for One Nation preferences in Labor-against-Coalition counts. This model adds Greens-against-One Nation and Others-against-One Nation, which are contest types the earlier models had no need to describe.

In Labor-against-Coalition final counts this model applies a national One Nation flow of 80 per cent to the Coalition and 20 to Labor, implemented by raising the Coalition's share of One Nation preferences by five points in every electorate, against 78.5 and four points in the previous model. The rural and regional figure in the Labor-preferences row is printed as 72–38 in the report, which sums to 110; it is given here as 72–28 and the point is recorded in the data notes.

Two movements between the models cut in opposite directions and both favour One Nation less than the headline vote does. Coalition preferences to One Nation in rural Labor contests fall from 79 per cent to 71, and Labor preferences to the Coalition in rural One Nation contests rise from 64 to 72. Each of those makes the final count harder for One Nation in exactly the seats it is winning, and the model still returns 60 seats for it, which says the primary vote is doing the work rather than the flows.

One flow crosses over, and it is the one with the least history behind it. In rural and regional Labor-against-One Nation contests, Others preferences went 53 to 47 in favour of Labor in the previous model and 52 to 48 in favour of One Nation in this one. That is a one-point movement across a line, in the category of voters whose second choice is hardest to predict, in the seats that decide the model's seat total. It is the single assumption on which most of section 01 turns.

The three-candidate step deserves naming too. Before any final count can be modelled the model has to decide which two candidates reach it, and this report does that by applying three-candidate preferred flows estimated from a mixture of 2025 election data and respondent-allocated data. That choice is upstream of every seat figure on this page and no uncertainty is published against it.