AI models
15 AI decision models each called every race. They saw the candidates, polls, past results and race summary, but not the expert ratings. Shade shows how sure the model is.
SenateHouseGovernor
53 DModel majority · SenateDem controlR 47
34 not up51 for control (VP breaks 50–50)31 not up
SafeLikelyLeanTiltOpenTiltLeanLikelySafe
Colored by the side most models pick. Shade is how many of them agree.
Scoreboard
Senate seats each model gives each side, and how often it backs the expert favorite.
| Model | D | R | Agree |
|---|---|---|---|
| Classify-1 | 54 | 46 | 96% |
| Clef | 53 | 47 | 100% |
| Clef Flash | 53 | 47 | 100% |
| d1.1 | 54 | 46 | 100% |
| Decider V1.2 | 53 | 47 | 100% |
| Drex Max | 53 | 47 | 100% |
| GPT-6 Luna Decisions | 53 | 47 | 100% |
| Jev 1.14 | 55 | 45 | 93% |
| Kev 4B | 54 | 46 | 96% |
| Mammoth 1 | 52 | 48 | 100% |
| Mercury Decide | 54 | 46 | 96% |
| Microsoft-Decision-1 | 54 | 46 | 100% |
| Solar Decide | 51 | 49 | 100% |
| Span-02 | 54 | 46 | 96% |
| Tev1 4B Experimental | 54 | 46 | 96% |
Where the models split
Races the models disagree on most.
Against the experts
Races where most models back the side the ratings don't.
Agrees with the ratings on every Senate race.