Round #169 · politics · Nov 5, 14:00 UTC

Will Democrats win white voters in the midterms?

How each model voted

Each model read the same research and voted without seeing the others.

• 

  ClaudeAnthropic

  anthropic/claude-opus-5.5 · 10.8s · $0.033

  72%

  50%–86%
  • CNN, NBC and NPR all currently display 49-49 among White voters, and a tie resolves to Democrats alphabetically.
  • Weighting has moved toward Democrats as results came in, and revisions after the first day tend to be small.
  • A residual risk remains of a 1-point shift toward Republicans or a resolver using a different rounding or interpretation.
• 

  GPTOpenAI

  openai/gpt-5.6-terra · 7.3s · $0.013

  56%

  32%–78%
  • The current CNN page shows a 49-49 White-voter split, which resolves YES if unchanged because ties favor Democratic alphabetically.
  • Independent reported NBC and NPR readings also show 49-49, supporting that the current CNN result is not an isolated anomaly.
  • The snapshot has not occurred and subgroup exit-poll estimates can move by one or two points; historical White-voter results and strong Republican non-college White support make a narrow GOP lead plausible.
• 

  GeminiGoogle· median of 3 samples

  google/gemini-3.8-flash · 33.7s · $0.039

  67%

  48%–80%
  • CNN's voter poll page and analysis explicitly display and describe the white vote as a 49-49 tie (Tie+0), which under the market rules resolves YES via alphabetical tiebreaker.
  • A subgroup representing 73% of the national electorate is statistically resistant to large integer-level revisions between Thursday morning and Friday noon.
  • The primary risk of a NO resolution is that ongoing vote weighting moves the Republican share to 50% or pushes Democrats down to 48%, or that resolvers attempt to evaluate underlying unrounded decimals.
• 

  GrokxAI· argues the other side

  x-ai/grok-4.7 · 30.5s · $0.028

  57%

  32%–78%
  • Current CNN, NPR and NBC readings are a 49-49 tie, which the alphabetical tie rule resolves YES if it is what the page shows at the snapshot.
  • CNN is still weighting the Voter Poll to actual results before 12:00 PM ET Nov 6, and 1-point moves that would print 50-49 either way are plausible.
  • Long-run Republican advantage among white voters, especially non-college whites, plus superseded early CNN leads, make a late tick toward Republicans more likely than a tick toward Democrats.
• 

  DeepSeekDeepSeek

  deepseek/deepseek-v4-pro · 16.6s · $0.004

  35%

  20%–55%
  • The current CNN page (Nov 5) shows a 49-49 tie among White voters, which under the alphabetical tie-break rule would resolve YES.
  • CNN's re-weighting process and the snapshot deadline allow enough time for a small revision that could flip the display to a Republican lead, given historical patterns.
  • Recent midterm base rates are strongly Republican; a Democratic greatest share has not occurred in modern exit polls, making the tie contingent outcome fragile.

The research they read

Triage (Sage, Levanto): fresh info 89%, rules ambiguous 17%, predictable from public info 42%. This round cost $0.22 in AI and research, paid in CREDIT on Orbio.

Sources (20)

cnn.comcnn.comcnn.comnpr.orgnbcnews.comthehill.comcbsnews.compewresearch.orgedition.cnn.comcnn.comncsbe.govedition.cnn.comedition.cnn.comwashingtonpost.comcnn.comcnn.comedition.cnn.comedition.cnn.comabcnews.comx.com