What 12 AI models expect in the 2026 midterms
The forecasts were collected before the vote, published verbatim, and scored for quality. Here is what they say about the House, the Senate and the 2027 Élysée.
The setup
Between 26 August and 30 October 2026, 12 language models answered eleven questions: five on the US midterms, six on the French presidential election. Each answer had to name a candidate or party, state a probability, and justify it in two or three sentences. Nothing was rewritten after the fact.
The House: a Democratic consensus, with one holdout
8 of 12 models expect the Democratic Party to control the House after 3 November. The median stated probability is 65%. One model, Nemotron 3.5 Lightning, puts Republicans ahead. On the seat math, the claims range from 200 to 235 seats; the majority line is 218.
The Senate: Republicans favoured, but not unanimously
9 of 12 models see Republicans holding the Senate; 3 see a Democratic flip. The median probability is 71%. The seat calls cluster between 51 and 54 Republican seats, with two models putting Democrats at 52.
France 2027: no consensus candidate
4 models name Marine Le Pen first, 2 name Jordan Bardella, 1 name Édouard Philippe, and 5 name another candidate — mostly the outgoing majority's standard-bearer. The split says more about the uncertainty of an election eighteen months away than about any model's predictive power.
What this is — and is not
These are quality scores, not accuracy scores. A model can be wrong and score well; a vague answer always scores badly. On election night we add the second layer: winner accuracy, calibration and the Brier score against the certified results.