Hysen Labsforecastbench-datasets
Forecastbench Datasets, updated nightly
ForecastBench Datasets ships nightly forecasting benchmarks
A repository of question and answer datasets used to evaluate how well language models forecast future events.
What it is
ForecastBench Datasets contains the datasets produced by ForecastBench, a benchmark that measures the forecasting ability of large language models. The repository is presented as the data companion to the broader effort, with the main explanatory site at forecastbench.org and the evaluation code in a separate forecastingresearch/forecastbench repository. The datasets are structured so that models can be scored on their ability to predict outcomes of future events rather than only answer static questions.
License and citation
The datasets are distributed under the CC BY-SA 4.0 license, and the README includes the full legal code link for that license. The accompanying research paper, authored by Ezra Karger, Houtan Bastani, Chen Yueh-Han, Zachary Jacobs, Danny Halawi, Fred Zhang, and Philip E. Tetlock, was presented at ICLR 2025 and is listed on arXiv as 2409.19839. A BibTeX entry is provided so that researchers can cite the benchmark in their own work.
Format and update cadence
The repository is formatted as a dataset card with the language set to English and the pretty name ForecastBench. The README notes that the datasets are updated nightly, which means the contents shift over time as new forecasting questions are added and resolved. As of the August 2026 snapshot the repository shows 36 stars and 3 forks, and the CC-BY-SA-4.0 license governs reuse and redistribution of the data.
Editorial conclusion
The datasets are released under CC BY-SA 4.0 and the linked paper appeared at ICLR 2025 under arXiv identifier 2409.19839. The repository records 36 stars and 3 forks as of August 2026 and is updated on a nightly schedule.
Community notes