SCIENCE · Settled
Will the highest temperature in London be 24°C on September 3?
The answer: NO
The panel of 14 models averaged 16% YES; 14 of 14 were on the right side. The crowd price when the panel was sealed: 15%.
THE BRIEF, WRITTEN AFTER THE OUTCOME
Panel and crowd correctly favoured NO; panel mean 16% vs crowd 15%
The resolved outcome was NO. The crowd price was 15% YES, the panel mean was 16% YES and the panel median was 13% YES. The best calls were the very low YES probabilities, led by Gemini 2.5 Flash at 5%, which matched the eventual NO. The misses were the higher YES forecasts, notably Kimi K2 at 38%, which overstated the chance of 24°C.
- Outcome: NO.
- Crowd (frozen Polymarket price) p_yes was 15%.
- Panel mean p_yes 16%, median 13%, range 5% to 38%.
- Top-performing forecasts were the lowest p_yes models (5, 10%).
- Largest miss was Kimi K2 at 38% YES, far above panel and crowd.
EVERY FORECAST, HIGHEST FIRST
- Kimi K238%
- Perplexity Sonar33%
- Claude Opus 4.122%
- Qwen3 235B18%
- Mistral Large18%
- Llama 4 Maverick15%
- DeepSeek V314%
- Claude Sonnet 4.512%
- GPT-510%
- GPT-5 Mini10%
- Grok 4.59%
- Gemini 2.5 Pro8%
- DeepSeek R16%
- Gemini 2.5 Flash5%