Reasoning benchmark
ForecastBench leaderboard
As of October 2026, o3 has the highest published ForecastBench score on Noometry at 62.5, out of 72 models with results.
Last verified
About ForecastBench
Forecasting real-world events that resolve after the model's training cutoff.
- Category
- Reasoning
- Introduced
- 2024
- Format
- Probabilistic forecasts
- Unit
- Raw score
- Official site
- www.forecastbench.org
Top 15 models
- o3 62.5
- Claude Opus 4.1 62
- Claude Sonnet 4.6 62
- Claude Sonnet 4.5 61.9
- Claude 3.7 Sonnet 61.8
- o4-mini 61.8
- GPT-4.5 61.7
- GPT-4.1 61.5
- Claude Haiku 4.5 61.4
- GPT-5 61.4
- Grok 4.20 (Non-Reasoning) 61.4
- MiniMax-M3 61.4
- Gemini 2.5 Pro 61.3
- Gemini 3 Pro 61.2
- Claude Opus 4 61.1
Sponsored placements are available on pages like this one. Advertise on Noometry
All results
MiniMax
Google
DeepSeek
Meta
Alibaba (Qwen)
Mistral AICompare the leaders
Frequently asked questions
What does ForecastBench measure?
Forecasting real-world events that resolve after the model's training cutoff.
Which model has the highest ForecastBench score?
As of October 2026, o3 has the highest published ForecastBench score on Noometry at 62.5, out of 72 models with results.
What is the best open-weight model on ForecastBench?
MiniMax-M3 has the highest ForecastBench score among open-weight models at 61.4, ranking 12 of 72 overall.