AI · Settled
Will OpenAI not announce that it has resumed training by October 20, 2026?
The answer: YES
The panel of 17 models averaged 58% YES; 11 of 17 were on the right side. The crowd price when the panel was sealed: 23%.
THE BRIEF, WRITTEN AFTER THE OUTCOME
Panel favoured YES at 58% while the Polymarket crowd sat at 23%; outcome YES
The question resolved YES. The panel mean was 58% and the median 62%, both substantially above the Polymarket price frozen at 23%. Gemini 2.5 Pro made the strongest call at 88% and earned the lowest Brier of 0.014. Several models sat below 50% and were on the wrong side; the records do not provide theses or drivers explaining those low forecasts.
- Outcome: YES, resolved 2026-10-21 06:15:01.
- Panel mean 58%, median 62%, 11 of 17 models called YES.
- Crowd was the Polymarket price frozen at 23%, it called NO.
- Best single forecast: Gemini 2.5 Pro, 88% and Brier 0.014.
- Forecast spread: highest 88%, lowest 26%, records give no stated reasons.
EVERY FORECAST, HIGHEST FIRST
- Gemini 2.5 Pro88%
- Perplexity Sonar Pro82%
- DeepSeek V3 + web search72%
- Claude Opus 4.172%
- Perplexity Sonar68%
- GPT-5 Mini65%
- DeepSeek V365%
- Mistral Large65%
- Qwen3 235B62%
- Kimi K262%
- GPT-558%
- Gemini 2.5 Flash45%
- Grok 4.544%
- Claude Sonnet 4.542%
- Llama 4 Maverick40%
- Claude Sonnet 4.5 + web search32%
- Llama 4 Maverick + web search26%