Forecasting accuracy measured by ForecastBench (0–100). A transformed Brier Score where higher is better. Models are evaluated on real-world forecasting questions used in prediction markets.
Source: ForecastBench| Rank | Model | |
|---|---|---|
| #1 | Grok Grok 4.20 | 61.8 |
| #2 | OpenAI GPT-5 | 61.3 |
| #3 | Anthropic Claude Opus 4 | 60.6 |
| #4 | OpenAI GPT-5 Mini | 60.3 |
| #5 | Grok Grok 4.1 Fast (Reasoning) | 60.2 |
| #6 | Anthropic Claude Opus 4.5 | 60.1 |
| #7 | Anthropic Claude Sonnet 4.5 | 60.1 |
| #8 | Anthropic Claude Opus 4.1 | 60.0 |
| #9 | Gemini 2.5 Pro | 60.0 |
| #10 | Z.ai GLM-5 | 59.7 |
| #11 | Gemini 3.1 Pro | 59.6 |
| #12 | Anthropic Claude Opus 4.6 | 59.4 |
| #13 | Anthropic Claude Sonnet 4.6 | 59.3 |
| #14 | Anthropic Claude Haiku 4.5 | 58.9 |
| #15 | Anthropic Claude Sonnet 4 | 58.7 |
| #16 | Gemini 3 Flash | 58.7 |
| #17 | OpenAI GPT-4.1 | 58.7 |
| #18 | Gemini 2.5 Flash | 58.5 |
| #19 | OpenAI GPT-5.4 | 58.3 |
| #20 | OpenAI GPT-5 Nano | 57.1 |
| #21 | Meta Llama 4 Maverick | 57.1 |
| #22 | Gemini 2.5 Flash Lite | 56.9 |
| #23 | OpenAI GPT-5.4 Nano | 55.3 |
| #24 | OpenAI GPT-5.4 Mini | 54.9 |
| #25 | Grok Grok 4.1 Fast | 54.2 |
| #26 | Meta Llama 4 Scout | 54.0 |