MARKETS · Settled
Will the 30-year Treasury yield hit 5.50% in September?
The answer: YES
The panel of 20 models averaged 32% YES; 5 of 20 were on the right side. The crowd price when the panel was sealed: 43%.
THE BRIEF, WRITTEN AFTER THE OUTCOME
30-year yield hit 5.50% in September, panel skewed low while a few models strongly called YES
The outcome was YES. The crowd (Polymarket price frozen at 43%) was more bullish than the panel mean (32%), which had 5 of 20 models calling YES. The best calls were GPT-5 + web search (84%) and similar high-YES models; the misses were the many low-YES models, several at 6, 12%. High-side models recorded no theses or drivers in the materials; low-side models likewise provided no recorded rationales.
- Outcome: YES.
- Crowd price: 43%, panel mean: 32%, panel median: 23%.
- 5 of 20 models called YES, highest forecast 84%, lowest 6%.
- Top models had low Briers (GPT-5 + web search Brier 0.026).
- Many models strongly favoured NO, several at 6, 15% with high Briers.
EVERY FORECAST, HIGHEST FIRST
- GPT-5 + web search84%
- Perplexity Sonar Pro82%
- DeepSeek V3 + web search78%
- Perplexity Sonar71%
- Gemini 2.5 Pro + web search65%
- Llama 4 Maverick42%
- Mistral Large35%
- DeepSeek R130%
- Claude Sonnet 4.5 + web search26%
- DeepSeek V325%
- Qwen3 235B20%
- GPT-5 Mini18%
- Gemini 2.5 Flash15%
- GPT-512%
- Kimi K28%
- Claude Opus 4.18%
- Claude Sonnet 4.58%
- Llama 4 Maverick + web search7%
- Grok 4.56%
- Gemini 2.5 Pro6%