AI · Settled
Will Google be the second-best AI lab at the end of September 2026?
The answer: YES
The panel of 20 models averaged 43% YES; 6 of 20 were on the right side. The crowd price when the panel was sealed: 84%.
THE BRIEF, WRITTEN AFTER THE OUTCOME
Panel split, crowd at 84% and the question resolved YES; panel mean was 43%
The market's frozen Polymarket price was 84% YES and the question resolved YES. The panel mean sat at 43%, with 6 of 20 models calling YES. GPT-5 + web search and the Baseline Copy Control matched the crowd at 84% and scored best, with GPT-5 + web search posting a Brier of 0.026. Most misses favoured NO or weak YES probabilities, the lowest forecast was 22% from Claude Sonnet 4.5. The panel therefore underweighted the outcome relative to the crowd.
- Outcome: YES, resolved 2026-09-30 18:15:02.
- Crowd: Polymarket price frozen at 84% YES, called it.
- Panel mean: 43% YES, 6 of 20 models called YES.
- Best individual forecast: GPT-5 + web search at 84%, Brier 0.026.
- Lowest forecast: Claude Sonnet 4.5 at 22%, Brier 0.608.
EVERY FORECAST, HIGHEST FIRST
- GPT-5 + web search84%
- Mistral Large62%
- Gemini 2.5 Pro + web search58%
- Llama 4 Maverick + web search56%
- DeepSeek R155%
- Llama 4 Maverick55%
- DeepSeek V3 + web search48%
- Claude Sonnet 4.5 + web search48%
- Gemini 2.5 Pro42%
- Grok 4.541%
- DeepSeek V340%
- Qwen3 235B38%
- GPT-5 Mini35%
- GPT-531%
- Kimi K228%
- Perplexity Sonar Pro28%
- Gemini 2.5 Flash28%
- Claude Opus 4.128%
- Perplexity Sonar27%
- Claude Sonnet 4.522%