← Map

2026 Midterms

Projected control

If every race broke the way it currently leans. Each seat is called from polling where it exists, then prediction markets, then how the state voted in 2024.

U.S. Senate

51 to control

51

Democratic

46

Republican

Democratic control by 5 seats · 3 independents

35 of 100 seats are on the ballot. The other 65 are held by senators not up for election this cycle and are counted at their current party.

19 from polling16 from 2024 result65 not up this cycle
10 seats within reach
  • Nebraska Senate0.0
  • Kansas Senate+0.9
  • Maine Senate+1.8
  • Iowa Senate+2.1
  • South Dakota Senate+2.3
  • South Carolina Senate+2.5
  • Ohio Senate+2.7
  • Michigan Senate+2.9
  • Texas Senate+2.9
  • Minnesota Senate+4.2

U.S. House

218 to control

196

Democratic

239

Republican

Republican control by 43 seats

All 435 seats are contested every two years, so nothing is carried over.

21 from polling63 from markets351 from 2024 result
20 seats within reach
  • AL-20.0
  • MT-1+0.2
  • FL-27+0.6
  • KY-6+0.8
  • FL-13+0.8
  • WI-2-0.9
  • WI-4-0.9
  • WI-5-0.9
  • WI-7-0.9
  • WI-8-0.9
  • NV-2+1.2
  • FL-22+1.3
  • MI-1-1.4
  • MI-2-1.4
  • MI-3-1.4
  • MI-5-1.4
  • MI-6-1.4
  • MI-8-1.4
  • MI-9-1.4
  • MI-10-1.4

What this is, and isn't

It is a seat count: for every race we ask who is ahead right now and add up the answers.

It is not a forecast. It carries no probability of control, no simulations, and no assumption that polling errors are correlated across states — which historically they are. A one-seat projected majority should be read as a coin flip, not a prediction.

Most House seats are never polled, so many are called from markets or from the 2024 presidential result. Those are weaker evidence than polling and are counted separately below each bar.

State legislatures

Not yet covered. State legislative races are polled so rarely, and the chambers are so numerous — 99 of them across 50 states — that any projection would be built almost entirely from the 2024 presidential result. That would look like data while carrying almost none, so it is left out until there is something real to show.