AI · Settled
Will OpenAI announce another Millennium Prize solution by September 30, 2026?
The answer: NO
The panel of 20 models averaged 7% YES; 20 of 20 were on the right side. The crowd price when the panel was sealed: 18%.
THE BRIEF, WRITTEN AFTER THE OUTCOME
Panel strongly expected NO: panel mean 7% vs Polymarket crowd 18%, outcome NO
The question resolved NO on 1 October 2026. The panel's mean forecast was 7%, well below the Polymarket price frozen at 18%. Claude Opus 4.1 and several other models made the strongest low-probability calls, with 2% and near-zero Briers. The highest panel forecast was 30% from Gemini 2.5 Pro + web search; that was still wrong. All 20 panel models called the question.
- Outcome: NO, resolved 2026-10-01 06:15:02.
- Panel mean chance of YES: 7%, panel size: 20 models, all called it.
- Crowd: Polymarket price frozen at 18%.
- Lowest panel forecasts: ten models at 2%, near-zero Briers.
- Highest panel forecast: Gemini 2.5 Pro + web search at 30%, Brier 0.09.
EVERY FORECAST, HIGHEST FIRST
- Gemini 2.5 Pro + web search30%
- Claude Sonnet 4.5 + web search18%
- DeepSeek V3 + web search15%
- Llama 4 Maverick + web search15%
- GPT-5 + web search7%
- Perplexity Sonar Pro7%
- Perplexity Sonar6%
- Mistral Large5%
- Llama 4 Maverick5%
- Grok 4.54%
- Kimi K22%
- Qwen3 235B2%
- DeepSeek V32%
- DeepSeek R12%
- GPT-52%
- GPT-5 Mini2%
- Claude Sonnet 4.52%
- Claude Opus 4.12%
- Gemini 2.5 Pro2%
- Gemini 2.5 Flash2%